Publication Record
Evidence current through: Thursday, November 19, 2026, at 1:15 a.m. UTC (8:15 p.m. ET and 5:15 p.m. PT on Wednesday, November 18).
Microsoft made Perception generally available at Ignite on November 17, 2026. Several related features are previews, and three Microsoft pages still describe Perception as a preview. The note reads each item at the status Microsoft gave it by the time above. Developments after that time are not reflected. Times in the text are given in UTC.
Research Note In Two MinutesNote synopsis · AI drafted · Analyst reviewed
Drafted by AI from the full research note and reviewed by a Greyhound Research analyst before publication. The analysis and conclusions are Greyhound Research’s own.
At Microsoft Ignite on November 17, 2026, Microsoft made Perception generally available. Perception is a team of AI agents that works inside Microsoft Defender, the company’s security console. Some agents sort and close security alerts, some investigate attacks, some probe the company’s own systems for weak spots, and some help fix them. Microsoft also widened its Sentinel SIEM, the system that collects security logs, for customers on its E5 and E7 suites, with longer retention and a lower price for added data. A new feature called vulnerability shielding, still in preview, puts temporary protection around systems that cannot yet be patched.
Greyhound Research’s position is that Perception is a serious step toward a security operations center (SOC) that runs at machine speed, and that its most important feature is also its main new risk. The Triage Agent closes alerts it judges to be false alarms without asking anyone first. Microsoft says it covers about 80% of the alerts in a typical Defender queue. When agents close the alerts, the audit becomes the control: someone the buyer trusts must regularly check a sample of what the agents closed. Greyhound Research reads this as a normal stage for a new kind of automation. The capability has arrived before the evidence of how often it is right, and buyers tend to do best when they use it from the first day and measure its decisions until the record supports lighter review.
Three findings carry the argument. First, the agents that defend the company are insiders too, with broad rights over the SOC itself, so they need the same narrow rights and reviews as any other agent. Second, Perception’s price has a meter, a cost estimator, and a calculator, but no published rate card, and it is priced separately from Security Copilot, whose capacity units many E5 customers already hold. Third, Microsoft’s own pages give three different versions of the Sentinel offer for E5 and E7 customers.
The note sets out a ten-move Greyhound CXO Runbook, with seven exhibits, seven tables, and six questions for Microsoft. None of the moves waits on Microsoft. The first three are simple to state. Audit a sample of agent-closed alerts every week. Give each security agent the narrow rights and reviews of a privileged account. Put an expiry date and a patch owner on every shield.
The evidence has limits. Microsoft’s product page and its documentation still describe Perception as a preview, so the note rests on the Ignite announcements for its status. No unit rate for Perception has been published, and the 80% figure measures coverage, not accuracy. Greyhound Research names the signals that would change its view within twelve months. Until they arrive, the advice holds: let the agents work the queue, and check their work.
In This Research NoteSection guide · AI drafted · Analyst reviewed
Eighteen sections in eight parts, each with a short summary. Select a title to go straight to that section. Each section opens with a Greyhound Spotlight, the one fact in it a reader would not expect, and thirteen end with a block named The Deeper Read, which holds the fuller argument and the passages it rests on. Select any line in that block to open its full text.
- Part 1What Ignite Changed
- 1.1What Ignite Changed: The SOC Gets A Workforce Of Agents
Perception went from first announcement to general availability in under four months. The change for buyers is that software now closes alerts, not only ranks them.
- 1.2What Microsoft Announced: General Availability, With Three Pages Still Saying Preview
Perception, seven of its agents, and the core of the integrated SOC console are generally available. Shielding, the full SIEM, and agent risk scoring are previews, and three Microsoft pages still call Perception a preview.
- Part 2Agents In The SOC
- 2.1The Triage Agent: Eight In Ten Alerts, With No Approval Before A Closure
The Triage Agent covers about 80% of a typical Defender queue and closes false alarms on its own. The 80% measures coverage, not accuracy, and no approval step sits before a closure.
- 2.2Agents As Insiders: The Defenders’ Agents Are Insiders Too
Microsoft calls agents the new insiders. Perception’s own agents hold some of the widest rights in the company, over alerts, investigations, and networks, and need the same narrow rights and reviews.
- 2.3Vulnerability Shielding: A Fast Shield Needs An Expiry Date
Shielding puts temporary protection, including network changes, around systems that cannot be patched yet. It is a preview, and the shield is only as good as the patch that follows it.
- Part 3The Money
- 3.1Pricing: A Meter And An Estimator, But No Rate Card
Perception is sold by consumption, on pay-as-you-go terms or an AI Flex Plan, and counts toward an Azure commitment. Rates vary by task and are not published. The only price tools are an estimator and a calculator.
- 3.2Security Copilot Commitments: Two Meters For One Alert Queue
Security Copilot runs on Security Compute Units, which E5 and E7 customers receive in part at no extra cost. Perception is priced separately. Microsoft has not said whether those units can pay for Perception.
- 3.3The Sentinel Entitlement: Three Microsoft Pages, Three Versions Of The Offer
E5 and E7 customers get a cheaper, longer-retention Sentinel SIEM inside Defender. Microsoft’s security blog, its Sentinel blog, and its documentation disagree on status, eligibility, and retention.
- Part 4The SOC And Its Rivals
- 4.1Staffing And Sourcing: Fewer Sorters, More Judges
Agents take over sorting, and Microsoft offers its own engineers and partners beside them. Analysts move to checking, detection, and response, and outsourcing contracts need new terms.
- 4.2Rivals Compared: Every SOC Vendor Now Sells Agents, On A Different Meter
Google, AWS, CrowdStrike, Palo Alto Networks, Splunk, and SentinelOne all sell agentic triage or investigation. Each counts usage in its own unit, so buyers have no common price to compare.
- Part 5The Risks And The Counter-Argument
- 5.1Risks To Watch: Where An Agentic SOC Can Go Wrong
Five risks, from unchecked closures to shields left in place, each with an early warning in the buyer’s own estate and an owner.
- 5.2The Strongest Objection: The Alert Queue Cannot Wait For An Audit
The fair counter-case: SOCs are overloaded, the agents are good enough, and Microsoft has approval controls. Most of it stands. The claim that the audit can come later does not hold.
- Part 6What To Do, By Role
- 6.1The Greyhound CXO Runbook: Ten Moves Before The Agents Run Unwatched
Ten moves for the SOC lead, the CISO, IT operations, procurement, finance, the CIO, and the board, each naming the assumption it corrects and who decides.
- 6.2Questions For Microsoft: What Buyers Will Ask Next
Six questions only Microsoft can answer, on rates, SCUs, the SIEM offer, triage error rates, shield expiry, and agent rights. Each is tied to a Runbook move.
- Part 7The Final Greyhound Standpoint
- 7.1Microsoft Perception: The Final Greyhound Standpoint
The agents work the queue, and the audit is the control. The limits of the evidence, and the signals that would change the view within twelve months.
- Part 8Appendix, Sources, And Disclosures
- 8.1Appendix: The Evidence Register, Source By Source
Every source behind this research note, with when it was read and where it is used. Passages quoted in The Deeper Read are not repeated here.
- 8.2Sources And References: Every Document Behind Every Claim
Thirty-nine sources in six classes, from Microsoft’s announcements and documentation to rival vendors, media, and the research this note builds on.
- 8.3Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright
Independence first, then the Financial Disclosure, intended use, research method and its limits, authoring, and copyright terms.
Where To Start, By RoleReader guide · AI drafted · Analyst reviewed
Find your role and read its sections in the order shown. Each line takes you straight to the section.
Chief information security officer
SOC lead
Head of procurement
Chief information officer
Head of IT operations
Chief financial officer
Part 1What Ignite Changed
What Ignite Changed: The SOC Gets A Workforce Of Agents
Greyhound SpotlightPerception went from first announcement on July 27 to general availability on November 17, 2026. That is sixteen weeks for a system that now closes security alerts without asking first.
For years, security tools have produced alerts and people have decided what to do with them. Ignite changed the second half of that sentence. Microsoft now sells a set of agents that decide, for a large share of alerts, whether they are real, and close the ones they judge to be false. A SOC is the team that watches a company’s systems for attacks. Its first line, often called tier 1, spends most of its day on exactly this sorting work.
The shift that matters is from advice to action. Earlier security automation, from correlation rules to playbooks, mostly ranked alerts or ran steps a person had approved in advance. Perception makes the call and acts on it, which moves the decision on a closed alert from a named analyst to a system the buyer configured. Most of Microsoft’s rivals already sell agents of this kind, so Greyhound Research expects the debate to move quickly from whether agents should decide to how their decisions are checked. The effect that is easy to miss sits in reporting. SOC metrics have measured how fast people clear the queue, and they will now need a measure of how well the agents clear it.
Exhibit 1 sets out Microsoft’s security calendar for 2026, from the launch of its E7 suite to Perception’s general availability. Each marked month carries a dated Microsoft announcement.
Greyhound Research · Exhibit 1 · Microsoft’s 2026 security calendar
From first announcement to general release in under four months
Dated Microsoft announcements on its security suite, Perception, and the integrated SOC console, January to December 2026.
Greyhound Research’s judgment: sixteen weeks is a short test. Treat the first six months of general availability as supervised use.
Exhibit 1 places Microsoft’s 2026 security announcements on a month grid, and shows Perception moving from first announcement in July to general availability in November. March 9: the E7 suite is announced. May 1: E7 is generally available at $99 per user. July 27: Project Perception is introduced. August 3: its limited, invitation-only preview opens. September 23: the integrated SOC console in Defender enters preview. October 19: Microsoft calls agents insiders. November 17: Perception is generally available and the SIEM offer for E5 and E7 customers is announced. January, February, April, and June are quiet, and December is ahead. A note beneath says AI Readiness Initiatives arrive in preview in January 2027.
The pace is the story. Microsoft announced Perception in July, opened a limited, invitation-only preview in August, put its new integrated SOC console into preview in September, and declared both ready in November. The customers invited to test it from August have had about three months of experience before the general release, and most buyers have had none. That is short for a system that will make decisions inside the SOC, and it explains why some of Microsoft’s own pages have not caught up.
Greyhound Standpoint:
At Greyhound Research, we believe Ignite moved agentic security from demonstration to purchase order. Perception’s agents now decide for a large share of the queue, in production, inside the Defender console that analysts already use. Microsoft deserves credit for that last point. A tool that sits where the work happens is adopted faster than one that asks analysts to change screens, and few customers could match this scale by hiring.
The difficulty is the calendar. Sixteen weeks from first announcement to general availability means few customers will have run the agents through a full quarter of their own alerts before signing. Enterprise software has a long record here. The first general release of a new class of product usually works, while the operating practice around it, the reporting, the exception handling, and the contract language, arrives a release or two later. Security orchestration tools followed a similar course, and most of their benefit came after teams redesigned their processes around them.
That history argues for patience, not refusal. Published error rates from production customers would shorten the case for caution. Until they exist, CISOs and SOC leads should treat the first six months of general availability as a supervised period, with a written plan for checking the agents’ decisions, and should bring that evidence to the board risk review before any reduction in staff or any change to an outsourcing contract.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1How did Perception reach general availability so quickly?
July to November. Hayete Gallot, Executive Vice President of Microsoft Security, introduced Project Perception on July 27, 2026, and wrote that it would enter public preview on August 3. Rob Lefferts, Corporate Vice President of Microsoft Threat Protection, announced the integrated SOC console in Defender as a preview on September 23. Microsoft’s getting-started guide describes that preview as limited and invitation-only. At Ignite, Satya Nadella said that Perception is now generally available, and the Book of News confirms the rename from Project Perception.
What changed since September. When the integrated console entered preview, Greyhound Research said that Microsoft had made integration easier to buy, and that the enduring test was whether customers retain the means to challenge and reverse what the agents do. Perception was then an invitation-only preview. General availability raises the stakes of that test rather than settling it, and this note applies it to the closures the agents now make on their own.
In the source’s own words
- The Official Microsoft Blog, Gallot on July 27enters public preview on August 3Source
- Microsoft Security Blog, Lefferts on September 23Integrated security operations center (ISOC) in Microsoft Defender is available in preview today.Source
- Microsoft’s Book of News, PerceptionMicrosoft Perception, previously known as Project Perception, is now generally availableSource
2What does Microsoft say the system is?
Teams of agents. Nadella described Perception as a complete agentic security system with red, blue, and green teams of agents that run through playbooks. Red agents simulate attacks, blue agents detect them, and green agents fix and remediate. Playbooks let customers compose teams of agents and trigger them to act on their own.
What the playbooks tell a buyer. Nadella’s description puts the weight on playbooks, the means by which customers compose teams of agents and set them to act on their own. That makes the playbook, more than any single agent, the unit a buyer governs. An agent can be reviewed once. A playbook that chains red, blue, and green agents decides which actions follow from which findings, so a change to it can change what the SOC does without any agent changing. Buyers should treat playbooks as change-controlled configuration, with an owner, a version history, and a review before each edit goes live.
In the source’s own words
- The Official Microsoft Blog, Nadella at IgnitePerception is a complete agentic security system with teams of red, blue and green agents that run through playbooksSource
- The Official Microsoft Blog, Nadella at IgniteWith Playbooks, you can compose teams of agents to tackle complex jobs and trigger them to take action autonomously.Source
Part 1What Ignite Changed
What Microsoft Announced: General Availability, With Three Pages Still Saying Preview
Greyhound SpotlightOn November 19, two days after Ignite, Microsoft’s older Perception product page still called it a preview, and two of its documentation pages still called it an invitation-only limited preview.
Microsoft announced a lot of security at Ignite, and the status of each item decides what a buyer can sign. Generally available means fully released, with support and service terms. A preview is a test release without them. Perception itself, most of its named agents, and the core of the integrated SOC console are generally available. Vulnerability shielding, the full SIEM inside that console, and risk scoring for agents are previews.
The pattern in that list is telling. The agents that act, by triaging, investigating, and prioritizing, are released. Two of the features that help a buyer watch and widen them, risk scoring for agents and the full SIEM inside the console, are still previews. In earlier platform launches the capability usually shipped ahead of the tools for governing it, and the gap tended to close over a release or two. Greyhound Research expects the same here, and Microsoft has signaled the direction by announcing agent risk scoring at all. Until it arrives, oversight of the released agents rests on the buyer’s own processes rather than on a Microsoft feature, and early plans should assume so.
Table 1 lists each security item by status, with what Microsoft has written and what it has not. ISOC stands for Integrated Security Operations Center, Microsoft’s name for the combined console.
Greyhound Research · Table 1 · Status ledger
Most of the agents are ready, and their outside data is not
Microsoft’s Ignite 2026 security items by status, with what Microsoft has written and what it has not, November 19, 2026.
Sign on the generally available rows. Treat every preview row, and every page still saying preview, as unsigned until confirmed in writing.
| Item | Status | What Microsoft has written | Not yet written |
|---|---|---|---|
| Microsoft Perception | Generally available | Red, blue, and green agents and playbooks in DefenderRenamed from Project Perception | A rate card. Three Microsoft pages still say preview |
| Blue team: Alert Triage, Attack Investigation, Detection Authoring, Threat Intelligence | Generally available | “about 80% of alerts in a typical Microsoft Defender queue”Coverage, not accuracy | Error rates for agent-closed alerts |
| AWS, Okta, and Salesforce signals for the agents | Preview | Third-party data for triage, investigation, and detectionMicrosoft pages differ on which sources triage covers | Release date, full source list |
| Red and green team: Recon, Posture Prioritization | Generally available | Find exposures and rank themWeb App Pen Test and Risk Reduction agents in preview | Default rights for each agent |
| ISOC core in Defender | Generally available | SIEM and threat protection in one consoleFull SIEM breadth in preview | One authoritative statement of the E5 and E7 terms |
| Vulnerability shielding | Preview | “targeted, temporary protection”Can change network accessibility | Duration, approval, reversal |
| Dynamic Agent Risk | Preview | Scores each agent’s exposure, blast radius, and behavior | Release date |
| AI Readiness Initiatives in Perception | Preview in January 2027 | Patching, authentication, and agent discovery tracks | Everything beyond the plan |
Table 1 lists eight Microsoft security items from Ignite 2026 by status, and shows Perception and most of its agents generally available while their third-party data and several features are previews. Generally available: Microsoft Perception, the four blue team agents with a claim of about 80% coverage of a typical Defender queue, the Recon and Posture Prioritization agents, and the core of the ISOC console. Preview: AWS, Okta, and Salesforce signals, vulnerability shielding, and Dynamic Agent Risk. Announced for January 2027: AI Readiness Initiatives. Each row names what Microsoft has not yet written, including a rate card, error rates, and the duration of shields. Statuses are as Microsoft published them through November 19, 2026.
The ledger has more generally available rows than most Ignite announcements, which is to Microsoft’s credit. Two problems remain. The connections to AWS, Okta, and Salesforce data, which make the agents useful beyond Microsoft’s own products, are previews. Three Microsoft pages also disagree with the announcement. Buyers should sign on the generally available rows and get the status of the rest in writing.
Greyhound Standpoint:
Greyhound Research’s view is that Perception is ready for production on Microsoft’s own signals and is still a preview everywhere else. That is a fair place for a first release. Most of the agents a SOC would start with, including triage, investigation, and posture prioritization, are generally available. The connections to AWS, Okta, and Salesforce are the parts still in preview.
Status matters because of how enterprise contracts work. An order form usually incorporates the vendor’s product terms and documentation by reference, so the page in force on the signing date becomes part of what was bought. Buyers who moved from perpetual licences to cloud subscriptions learned that the terms behind a link can change during a contract, and that disagreements tend to turn on which version applied. Three Microsoft pages that still say preview create that kind of ambiguity. The likely cause is the ordinary lag between a launch and its documentation, not a change of plan, and if Microsoft updates all three in the coming weeks this concern largely falls away.
Procurement leads should ask Microsoft, before signature, to correct the older product page and the two Learn pages, and to list in the order form which agents and data connections are generally available on the signing date. Any connection still in preview belongs in a pilot schedule with its own exit terms, reviewed again at the first renewal.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1Which agents are generally available, and which are previews?
By team. Aleš Holeček, President and Chief Architect of Microsoft Security, lists four blue team agents as generally available: Alert Triage, Attack Investigation, Detection Authoring, and Threat Intelligence. On the red team, the Recon Agent is generally available and the Web App Pen Test Agent is a preview. On the green team, the Posture Prioritization Agent is generally available and the Risk Reduction Agent is a public preview. Signals from AWS, Okta, and Salesforce are in preview, according to the same post.
Other previews. Dynamic Agent Risk, which scores the risk of each agent in the company, is a preview. AI Readiness Initiatives in Perception arrive in preview in January 2027.
In the source’s own words
- Microsoft Security Blog, Holeček at Ignitefour blue team agents, now generally available: Alert Triage Agent, Attack Investigation AgentSource
- Microsoft Security Blog, Holeček at Ignitethe Recon Agent, now generally available, and the Web App Pen Test Agent, now in previewSource
- Microsoft Security Blog, Holeček at IgnitePosture Prioritization Agent, now generally available, and the Risk Reduction Agent, now in public previewSource
2Which pages disagree?
Three stale pages. Microsoft’s older Project Perception product page still describes a workforce of specialized AI agents now in preview, although the Security Copilot pricing page now links to the newer one. Its Learn guide to getting started, last updated September 6, says Perception is in a limited public preview open to a small, invitation-only set of customers, and the Triage Agent’s Learn page, last updated August 19, carries the same notice. The newer product page carries the Microsoft Perception name and states no status. Across the whole event, 76 of 215 announced features were generally available on announcement day, by one count of every Ignite feature by status.
What three stale pages tell a buyer. The disagreement is most likely a documentation lag, and the newer product page already carries the new name. It still matters, because the teams that configure the product read the Learn pages, not the launch post. A setup guide written for an invitation-only preview may also describe preview defaults and limits. Until the pages are updated, buyers should treat the launch post as the statement of status and the Learn pages as the statement of behavior, and check that the two agree for each agent they switch on.
In the source’s own words
- Microsoft Learn, Get started with Project PerceptionLimited Public Preview Program means Perception is available to a small, invitation-only set of customers for a defined preview window before broader public availability.Source
- Microsoft Learn, Triage AgentProject Perception is currently in Limited Public Preview.Source
- Ignite feature count by status, November 18Only 76 of them (35%) are generally available today.Source
Part 2Agents In The SOC
The Triage Agent: Eight In Ten Alerts, With No Approval Before A Closure
Greyhound SpotlightWhen the Triage Agent decides an alert is a false alarm, it closes it. Microsoft’s documentation describes no approval step before that happens.
The Triage Agent reads each new alert, gathers the evidence, and decides whether the alert is malicious or a false alarm. If it is malicious, the agent leaves the incident open for an analyst. If it is a false alarm, the agent marks it as such and resolves it. Microsoft says the agent covers about 80% of the alerts in a typical Defender queue, based on its own analysis of about 8 million alerts. That figure says how many alerts the agent can handle. It does not say how often the agent is right.
What is new here is not automation but unreviewed closure. Security teams have long suppressed known noise with tuning rules, and Microsoft’s setup switches off the built-in rules the agent replaces. The difference is that a rule does the same thing every time and can be read before it runs, while an agent weighs evidence and may decide two similar alerts differently. An agent can only be judged on its record. Greyhound Research expects accuracy figures to appear in time, because buyers will ask for them at renewal. Less discussed is the effect on detection quality. Alerts the agent closes quietly may no longer prompt anyone to fix the noisy detection that raised them.
Exhibit 2 follows one alert through the Triage Agent, from arrival to closure, and marks where a person is involved. A false positive is an alert that turns out to be harmless.
Greyhound Research · Exhibit 2 · One alert through the Triage Agent
A harmless verdict closes the alert without a person reviewing it
The path of one Defender alert through Perception’s Triage Agent, read against Microsoft’s documentation, November 19, 2026.
The harmless verdict is the unreviewed one. Sample those closures every week, because the agent’s own explanation is not a check.
Exhibit 2 follows one alert through the Triage Agent in five numbered steps, and shows that a person is involved only when an alert is judged malicious or when an analyst chooses to give feedback. Step one, arrive: the agent runs automatically under its own Entra Agent ID. Step two, gather: the agent reasons over evidence and writes an explanation. Step three, malicious: a true positive stays open for an analyst. Step four, harmless: the agent classifies a false alarm as a false positive and resolves it, with no approval step stated. Step five, learn: feedback teaches the agent only when someone chooses, and for email alerts only. The steps and grades are Greyhound Research analysis.
The person appears at only two points: when an alert is judged malicious, and when an analyst chooses to give feedback. A harmless verdict closes the alert without a person reviewing it. The agent writes an explanation for each verdict, which helps. But Nadella made the deeper point himself at Ignite: what a model says about itself is not independent evidence. An explanation written by the agent that made the call is not a check on the call.
Greyhound Standpoint:
Greyhound Research holds that the Triage Agent is the most useful thing Microsoft shipped for the SOC at Ignite, and that its closures need an audit the agent does not perform on itself. An agent that handles about eight in ten alerts in a typical Defender queue returns analyst hours to investigation and response.
The risk follows from arithmetic rather than from any flaw in the agent. A system that is usually right and closes alerts unseen will, across millions of decisions, eventually close a real attack, and the first sign may be the attacker’s next move. The agent’s explanation of each verdict helps an analyst understand the call, but does not test it. Card fraud scoring and financial audit reached the same answer long ago: a random sample of decisions, re-examined by a person independent of the system. Greyhound Research wrote in May 2025 that security maturity would be judged by autonomy, not headcount. With agents now closing alerts at scale, the view has moved one step: autonomy remains the goal, and a measured error rate is how it is earned.
Until Microsoft publishes error rates by alert type, SOC leads should sample the alerts the agent closed every week, have an analyst re-examine them and track the share that should have stayed open. That share, not the 80% coverage figure, belongs in the monthly report to the CISO and in the service levels of any outsourcing contract at renewal.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What exactly does the agent do on its own?
Closes, then explains. Microsoft’s Learn page for the Triage Agent says the agent runs automatically when a relevant alert is created, resolves false alarms as false positives, and leaves true positives open for an analyst. It describes no approval step before a closure. Feedback from analysts is not used to retrain the agent until someone explicitly chooses to teach it, and that learning applies only to email and collaboration alerts. The agent needs its own Microsoft Entra Agent ID, which the setup wizard creates. Setup also switches off the built-in tuning rules that already resolve supported alert types, so the agent takes over closures that those rules used to make.
In the source’s own words
- Microsoft Learn, Triage AgentIf the alert is determined to be a false alarm, the agent classifies it as a False Positive and resolves it accordingly.Source
- Microsoft Learn, Triage AgentThe agent needs a Microsoft Entra Agent ID to work.Source
- Microsoft Learn, Triage AgentThe setup process disables built-in tuning rules that resolve supported alert types.Source
2What does the 80% figure measure?
Coverage, with a caveat. The Book of News says the agent covers about 80% of alerts in a typical Defender queue, and handles AWS and Salesforce alerts and incidents. Microsoft’s Perception blog on Tech Community bases the figure on its analysis of about 8 million alerts, and its footnote warns that the figure is not a guaranteed rate for any one customer. The same post lists Okta and Salesforce, not AWS, as triage sources, so the two Microsoft pages differ on which outside data the agent triages. Neither page gives an accuracy figure. For its earlier Security Copilot triage agent, Microsoft’s Learn page says that agent has shown measurable improvements in triage accuracy and efficiency in controlled evaluations, without giving figures there.
What a customer says. Karsten Bode, Enterprise IT Security Architect at TÜV SÜD, is quoted in Microsoft’s Perception blog saying that the Triage Agent covers many incidents and makes the right threats available to analysts. That is credit for reach. Neither he nor any other customer on the pages read publishes a share of closures that were later reversed.
What the two Microsoft pages tell a buyer. The figure is offered with care. Microsoft states its basis and warns that it is not a guarantee for any one customer, which is to its credit. Read alongside the difference over which outside sources the agent triages, it tells a buyer that coverage depends on what data is connected, and that a company’s own coverage may sit well above or below the typical queue. The customer quotation praises reach, not accuracy. A buyer’s first measurement should therefore be its own coverage by alert source, taken before the accuracy sample begins, so that both numbers describe the same queue.
In the source’s own words
- Microsoft’s Book of News, PerceptionThe Triage Agent now covers about 80% of alerts in a typical Microsoft Defender queue.Source
- Microsoft Tech Community, Perception blogis not a guaranteed triage rate for an individual customer.Source
- Microsoft Tech Community, Perception blog, Karsten Bode of TÜV SÜDThe Triage Agent covers many incidents, performing analyses and making the right threats available to analysts.Source
- Microsoft Learn, Security Alert Triage Agenthas demonstrated measurable improvements in triage accuracy and efficiency in controlled evaluationsSource
- The Official Microsoft Blog, Nadella at IgniteAnd what the model tells us about itself is not independent evidence.Source
Part 2Agents In The SOC
Agents As Insiders: The Defenders’ Agents Are Insiders Too
Greyhound SpotlightMicrosoft’s setup guide tells customers to give each Perception agent’s identity access to every scope it needs. Anyone with the Security Reader role can then use all the configured agents.
Microsoft’s security leaders call AI agents insiders, and Greyhound Research agrees, as its companion note on Azure agent infrastructure argues. That label applies to Perception’s own agents as much as to any sales or finance agent. An agent that can close alerts, read every investigation, or change what a machine can reach on the network holds power over the SOC itself. If such an agent is misled by what it reads, or holds more rights than its task needs, it can narrow what the team it serves is able to see.
Security teams tend to treat their own tools as trusted infrastructure, and that habit needs revisiting. For decades the security stack sat largely outside the access reviews it ran on everyone else, because its accounts belonged to the people doing the reviewing. An agent is different, because it acts on what it reads, including material an attacker can shape, such as an email or a log entry. Greyhound Research expects agent rights to become a standing item in identity reviews within a year or two, much as service accounts did. The quieter consequence is for separation of duties. When one agent writes detections and another closes the alerts they raise, no person may see the whole loop.
Exhibit 3 sets Perception’s agents against five kinds of power they may hold, as Microsoft’s documents describe them. Read access means the agent can see data. Act means it can change something.
Greyhound Research · Exhibit 3 · The defenders’ agents as insiders
Two powers change the SOC without a person approving
Five powers Perception’s agents may hold, against four agents or features, as Microsoft’s documents describe them, November 19, 2026.
Review the powers that act before the powers that read. Govern each agent as a privileged account, starting with the two that need no approval.
Exhibit 3 is a matrix of five powers against four Perception agents or features, and shows two powers that act without a stated approval: the Triage Agent closing alerts, and shielding changing protections on live systems, shown as tinted rows. All four read alerts, incidents, and evidence. Triage, Investigation, and Detection Authoring read AWS, Okta, or Salesforce data in preview. Detection Authoring writes detection rules that analysts review before deployment. Other cells are not described in Microsoft’s documents. The marks are Greyhound Research analysis, not measured data.
Every agent reads widely, which is what makes it useful. Three of the five powers change something: closing alerts, writing detections, and changing protections on live systems. Those are the powers an insider review should look at first. Microsoft has written controls that help, including least-privilege access, approval controls, and audit logs. The grades show where the buyer has to set them, because a default is not a decision.
Greyhound Standpoint:
In Greyhound Research’s assessment, the agents defending the company are insiders too, and should be governed as privileged accounts. An agent that can close alerts, write detections, or change what a machine can reach on the network shapes what the SOC sees. Misled by what it reads, or holding more rights than its task needs, it can narrow what the SOC sees.
Service accounts and administrator accounts have long gathered rights that outlive the project that created them, which is why privileged access management became a standard discipline. Microsoft’s setup guide asks customers to grant each agent identity every scope it needs, and lets any Security Reader use all configured agents. That is a reasonable starting default, and Microsoft deserves credit for building least-privilege access, approval controls, and audit logs into the product. A default is still not a decision. This matches the companion Greyhound Research note on Azure agent infrastructure: isolation is the floor, and narrow rights, approval, and identity are what stop misuse.
CISOs should give each Perception agent an accountable owner, a written list of the actions it may take alone, approval before any change to a live system, and a quarterly review that removes unused rights, and should limit who may start an agent. Watching agents without trimming their access records incidents more than it prevents them. A narrower Microsoft default would ease the work, and the first review belongs in the next identity governance cycle.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What rights do the agents start with?
Broad by design. Microsoft’s getting-started guide says that where access is scoped by device groups or role-based rules, customers must grant the agent identity access to all scopes the agent needs. It says users with the Security Reader role can use all configured agents, and that only Security Admins can set them up or remove them. The Triage Agent’s page asks customers to give the agent’s identity only the permissions it needs, which is the right advice, and lists the rights it requires for each alert type, including the right to manage alerts. Holeček writes that the customer’s team sets the guardrails and signs off on high-impact actions.
What the setup guidance tells a buyer. The two Learn pages pull in different directions, and both are reasonable. One asks for every scope the agent needs, so that it works across the estate. The other asks for only the permissions each task needs. In practice the pressure of setup tends to win: the first goal is to make the agent work, and narrowing comes later, if it comes. The Security Reader rule adds a second point. Read access to a security console is commonly granted widely, so the group able to use the agents may be larger than the one a CISO would choose. Both are settings a buyer can review on the day of setup.
In the source’s own words
- Microsoft Learn, Get started with Project Perceptionyou must grant the agent identity access to all scopes required for the agent to operate.Source
- Microsoft Learn, Get started with Project PerceptionUsers with the Security Reader role can use all configured agents.Source
- Microsoft Security Blog, Holeček at IgniteYour team sets the guardrails and signs off on high-impact actionsSource
- Microsoft’s Book of News, PerceptionLeast-privilege access, approval controls, audit logging, usage views and a cost calculatorSource
2Where does the insider framing come from?
Microsoft’s own words. Holeček wrote at Ignite that agents are the new insiders. Gallot wrote on October 19 that agents inside companies are agentic insiders and are frequently overprivileged. CrowdStrike, a rival, announced detection for what it calls the agent insider threat on November 4. Greyhound CIO Pulse 2025 found that 69% of CIOs named silent privilege creep as their top concern with AI agents, and that 57% of CISOs in Global 2000 firms cited agent identity ambiguity as a top-five security concern for GenAI deployment.
In the source’s own words
- Microsoft Security Blog, Holeček at IgniteAgents are the new insiders.Source
- The Official Microsoft Blog, Gallot on October 19These agents are frequently overprivileged.Source
- CrowdStrike, November 4The hardest case is the agent that is manipulated or drifts from its intended behavior and never breaks a rule.Source
Part 2Agents In The SOC
Vulnerability Shielding: A Fast Shield Needs An Expiry Date
Greyhound SpotlightMicrosoft’s shielding can change what a vulnerable machine can reach on the network, on its own, during an attack. Microsoft calls the protection temporary and has not said who decides when it ends.
When a serious flaw is published, most companies need days or weeks to patch every affected system. Attackers often move faster. Vulnerability shielding, a preview announced at Ignite, fills that gap. Defender places a targeted, temporary control around a vulnerable machine, such as cutting off some of its network access, until the patch arrives. It is a sound idea, and it is also an agent changing live systems.
Shielding is a familiar idea in an unfamiliar form. Network teams have long placed temporary blocks around exposed systems, but a person usually proposed each change and a change board approved it. Here the control is placed automatically during an attack, which is when speed matters most and review is hardest. For buyers, that moves the important decision from the moment of the shield to the policy written beforehand: which systems may be shielded without approval, and which may not. Greyhound Research expects shielding to prove most useful on systems that cannot be patched quickly, such as older servers. A consequence that gets little attention falls on business owners, who may see an application lose network access without knowing why.
Exhibit 4 follows one flaw from the day it is published to the day its shield is removed. The bar shows the stages, and the strip beneath shows who owns each one.
Greyhound Research · Exhibit 4 · The life of a shield
Microsoft writes the shield, and the patch is left to the customer
One vulnerability from disclosure to closure, marked where Microsoft’s Ignite pages describe the work and where they stop, November 19, 2026.
Microsoft has automated the fast half. A shield without an expiry date and a patch owner can become a permanent rule with no owner.
Exhibit 4 follows one vulnerability along a bar from disclosure to closure in five stages: flaw published, exposure assessed, shield applied, patch tested and deployed, and shield removed. A marker shows where Microsoft’s part ends, after the shield is applied. Assessing and shielding are written by Microsoft, in preview, with targeted, temporary controls that can include network changes. Patching and removing the shield are left to the customer, with no duration, approver, or reversal documented. Stage lengths are illustrative. The mapping is Greyhound Research analysis.
Microsoft has written the first half of the path: assess exposure, then shield. The second half belongs to the customer. Someone has to patch the machine, and someone has to remove the shield. A shield left in place can turn into a permanent network rule whose purpose is soon forgotten. A shield that ends before the patch is done leaves the machine as open as before.
Greyhound Standpoint:
Greyhound Research regards vulnerability shielding as the right answer to a real gap, and as a preview whose value depends on the patch that follows it. The gap is old. Attackers often move on a published flaw within days, while large estates take weeks to patch. Placing a temporary control around an exposed machine, often called virtual patching, has been common practice with intrusion prevention systems and web application firewalls for years. Microsoft’s contribution is to make that step automatic and to tie it to Defender’s view of exposure, which could shorten the riskiest window.
The weak point is at the other end of the shield’s life. Temporary firewall exceptions that turn into permanent ones are among the most familiar findings in infrastructure audits, because adding a rule is easy and removing one requires someone to own it. Greyhound Research made the related point about Oracle in October: when AI finds flaws faster, the scarce resource is the capacity to fix them. Documentation from Microsoft on shield duration, approval, and reversal would answer much of this concern.
Before turning shielding on, CISOs and heads of IT operations should agree who approves a shield on a critical system, how long a shield may last, and who owns the patch. Every shield should carry an expiry date and a named patch owner in the change log, and the change advisory board should review open shields every month.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What does Microsoft say shielding does?
Temporary, targeted, autonomous. Holeček describes vulnerability shielding as a preview that applies targeted, temporary protection to vulnerable assets, and says that in supported scenarios the protection can include changes to network accessibility. The Book of News says the capability works by autonomously deploying targeted, temporary controls to known vulnerable systems during an active attack. Neither page says how long a shield lasts, who approves it, or how it is reversed.
In the source’s own words
- Microsoft Security Blog, Holeček at IgniteVulnerability shielding, available in preview, helps protect vulnerable assets from exploitation by applying targeted, temporary protectionSource
- Microsoft Security Blog, Holeček at IgniteIn supported scenarios, these protections can include adjustments to network accessibilitySource
- Microsoft’s Book of News, Defenderautonomously deploying targeted, temporary controls to known vulnerable systemsSource
2How does it relate to what Defender already does?
A cousin of predictive shielding. Defender already offers predictive shielding in preview, which, according to Microsoft’s documentation, contains at-risk user accounts, hardens group policy settings, or restricts access to sensitive data during an ongoing attack, and needs a Defender for Endpoint license. Perception’s AI Readiness Initiatives, due in preview in January 2027, include a patching initiative that will assess, shield, and coordinate patching for priority vulnerabilities.
Why one containment policy is enough. Microsoft is building a family of automatic containment features rather than a single one. Predictive shielding acts on accounts, policy, and data access. Vulnerability shielding acts on machines and their network reach. The patching initiative due next year would join shielding to patching. For a buyer, that points to one policy covering all of them: which assets may be contained automatically, for how long, and who is told. Writing that policy once, before the features multiply, is simpler than writing a rule for each feature as it ships.
In the source’s own words
- Microsoft Learn, Predictive shieldingPredictive shielding (Preview) is a proactive defense strategy designed to anticipate and mitigate threats as part of an ongoing attack.Source
- Microsoft’s Book of News, AI Readiness InitiativesPatching and vulnerability management: Will assess, shield and coordinate patching for priority vulnerabilities.Source
Part 3The Money
Pricing: A Meter And An Estimator, But No Rate Card
Greyhound SpotlightPerception’s spending counts toward a customer’s Azure consumption commitment. A SOC agent can draw down money set aside for cloud computing without anyone in the cloud team approving it.
Most security software is sold per user or per device, so the bill is known in advance. Perception is sold by consumption: the customer pays for the work the agents do. Microsoft says agents use capacity at different rates depending on how hard the task is. A customer can pay as it goes, or commit in advance through an AI Flex Plan of one or three years. Its spending also counts toward a Microsoft Azure Consumption Commitment, or MACC, the multi-year spending promise many large companies already hold.
The change in buying model is larger than it looks. Security budgets have rested on per-user and per-device counts that can be forecast a year ahead. A consumption meter tied to alert volume moves with the threat environment instead, so a bad month for attacks can be a bad month for the budget. Greyhound Research expects many buyers to ask for committed volumes with a ceiling, as they did when computing moved to consumption, and expects Microsoft to publish more pricing detail as the product matures. The less visible effect sits in incident response. The agents will work hardest during a major attack, when the team has least time to watch spending, so any cap must be set beforehand.
Exhibit 5 sorts what Microsoft has published on Perception’s price into three groups: what is written, what the buyer must set, and what is not stated. A rate card is a published list of unit prices.
Greyhound Research · Exhibit 5 · What is known about the price
A way to buy and a way to watch, but no price
Microsoft Perception’s pricing as published, the terms the buyer must set, and what is not stated, November 19, 2026.
An estimator is not a rate card. Make no AI Flex Plan or MACC commitment until the unit and its price are in writing.
Exhibit 5 groups what is known about Perception’s price into three cards. Written by Microsoft: consumption pricing with rates that vary by task, purchase by pay-as-you-go or an AI Flex Plan of one or three years, and a cost estimator, usage dashboards, and a calculator. For the buyer to set: a monthly budget, caps for each agent and month, and who may let Perception draw on the Azure commitment. Not stated by Microsoft: what a unit is, what it costs, and whether Security Copilot capacity can pay for it. The grouping is Greyhound Research analysis.
Microsoft has written how to buy, and given the buyer tools to watch and budget: a cost estimator, usage dashboards, and a calculator. What it has not published is the price of a unit, or what a unit is. A buyer can estimate, but cannot check the estimate against a list. For a system that runs on its own whenever an alert arrives, that turns the SOC’s alert volume into the security budget’s main variable.
Greyhound Standpoint:
Greyhound Research’s judgment is that consumption pricing suits agentic work, and that a meter without a rate card is not yet a price. It fits agents whose load follows the alert queue. Microsoft has also given buyers good tools to watch spend: an estimator at setup, usage dashboards, and a calculator. What is missing is the unit and its rate, the numbers needed to compare Perception with hiring, with outsourcing, or with a rival.
Perception runs whenever an alert arrives, so its cost follows alert volume, which the buyer does not control. Its spending also counts toward the Azure consumption commitment, so a security workload can draw down funds the cloud team planned for other uses. Early cloud bills arrived before the habits of governing them, and FinOps grew up to close that gap. The lesson was not to avoid consumption pricing but to set budgets, owners, and caps before the meter starts. Greyhound CIO Pulse 2026 finds that 84% of more than 1,000 enterprise technology leaders have moved from AI adoption to AI accountability, and an unpriced meter is hard to account for. A published rate card with caps would change this view quickly.
Procurement should obtain unit rates in writing before any AI Flex Plan or MACC commitment, ask for a cap per agent and per month, and agree with the CFO, at the next commitment review, who may approve Perception drawing on the Azure commitment.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What does Microsoft publish on Perception’s price?
How to buy, not what it costs. Microsoft’s Perception product page says Perception uses consumption-based pricing, bought pay-as-you-go or through the AI Flex Plan, and that this usage counts toward an existing MACC. It describes a cost estimator available at setup and at any time, and a usage dashboard in the Defender portal. The older Project Perception page says agents consume capacity at different rates depending on the intensity of the task. Neither page names a unit or a rate.
In the source’s own words
- Microsoft Perception product page, FAQMicrosoft Perception uses consumption-based pricing, with two ways to purchase:Source
- Microsoft Perception product page, FAQyour pay-as-you go and AI Flex Plan usage counts toward it.Source
- Microsoft Perception product page, FAQPerception comes with a cost estimator tool that customers can access at set up or at any time while using the product.Source
- Project Perception product page, FAQagents' consumption is measured at different rates depending on the intensity of the task they performSource
2What do the Ignite pages add?
Dashboards and a calculator. Holeček writes that new cost management dashboards and a cost estimator help customers track AI usage and forecast spending. The Book of News lists usage views and a cost calculator among Perception’s controls, and Microsoft’s Perception blog says usage dashboards and a cost calculator give visibility into agent activity and spending. Google, by comparison, says its security operations product will let customers set token budgets for each agent.
Watching spend is not pricing it. The tools Microsoft describes are good at telling a buyer what it has spent and, through the estimator, what it might spend. They do not tell a buyer what a unit should cost, which is the number procurement needs to negotiate and finance needs to approve. Google’s per-agent token budgets point to one direction the market may take: controls that cap spend for each agent, not only views that report it. Buyers can reasonably ask Microsoft for the same, and should test the estimator against a month of actual usage before relying on it in a business case.
In the source’s own words
Part 3The Money
Security Copilot Commitments: Two Meters For One Alert Queue
Greyhound SpotlightMicrosoft sells two triage agents for Defender. One runs on Security Copilot capacity, which E5 customers receive in part at no extra cost. The other is Perception’s, on a separate meter.
Many large customers already pay for AI in the SOC through Security Copilot, Microsoft’s security assistant. It is billed in Security Compute Units, or SCUs, a block of computing capacity bought by the hour. Eligible Microsoft 365 E5 and E7 customers receive 400 SCUs a month for every 1,000 user licenses, up to 10,000 a month, with the rollout still phased. Microsoft’s Perception page says the two products work together and are priced separately. Nothing Microsoft has published says whether SCUs, bought or included, can pay for Perception’s agents.
The split between the two meters says something about direction. Microsoft describes Security Copilot as AI that assists and Perception as AI that acts, and its announcements now concentrate on the agents that act. Greyhound Research reads that as a sign that new agent work is likely to land on the Perception meter, while Security Copilot keeps the assisting role. That is an expectation, not a statement Microsoft has made. For a buyer, the practical effect is that included SCUs may fund less of the SOC’s agent work over time than planned. Budget owners would be wiser to treat the E5 allowance as capacity for defined tasks, not free AI capacity for the SOC.
Table 2 sets the two meters side by side. Provisioned capacity is capacity reserved and paid for in advance. Overage is usage beyond it, billed when it happens.
Greyhound Research · Table 2 · Two meters, side by side
Two triage agents for Defender, billed on two meters
Security Copilot and Microsoft Perception compared on how each is metered, included, and bought, November 19, 2026.
One queue on two meters risks paying twice. Map each alert type to its agent and its meter before renewing either.
| Term | Security Copilot | Microsoft Perception |
|---|---|---|
| What it is | Generative AI assistant: “AI that assists” | Agentic system: “AI that acts” |
| Unit | Security Compute Unit, provisioned by the hour, with overage on demand | Not named Consumption at rates that vary by task |
| Included with E5 and E7 | Yes 400 SCUs a month per 1,000 user licenses, up to 10,000Phased rollout | Not stated |
| How it is bought | Provisioned capacity billed monthly, plus overage | Pay-as-you-go or AI Flex Plan, counted toward an Azure commitment |
| Its triage agent | Security Alert Triage AgentEmail generally available, cloud and identity in preview. Needs provisioned SCUs | Triage AgentGenerally available. No SCU prerequisite stated |
| Published rate | A worked example uses $4 a provisioned SCU and $6 an overage SCU | Not stated |
| Can one pay for the other | Not stated | Not stated Microsoft says the two are priced separately |
Table 2 compares Security Copilot and Microsoft Perception on seven rows, and shows two triage agents for Defender on two meters, with nothing published on whether one can pay for the other. Security Copilot is AI that assists, billed in Security Compute Units provisioned by the hour with overage, and eligible E5 and E7 customers receive 400 SCUs a month per 1,000 user licenses, up to 10,000, in a phased rollout. Its Security Alert Triage Agent needs provisioned SCUs. Perception is AI that acts, sold by consumption at rates that vary by task, through pay-as-you-go or an AI Flex Plan, with no unit or rate published and no stated E5 or E7 inclusion. Its Triage Agent is generally available. Microsoft says the two are priced separately.
The two products now overlap where it matters most to a SOC budget. Security Copilot’s own Security Alert Triage Agent, part generally available and part preview, requires provisioned SCUs. Perception’s Triage Agent does similar work on Perception’s meter. A customer could pay twice for the same queue, or hold SCUs that its agents no longer use. Microsoft has not said whether either agent will be retired.
Greyhound Standpoint:
Greyhound Research’s reading is that Perception has made Security Copilot commitments a planning question that Microsoft has not yet answered. Customers who bought provisioned SCUs, or who counted on the E5 and E7 allowance of 400 SCUs a month for every 1,000 user licenses to fund agents, may find that the triage agent they now want runs on a different meter. Microsoft has been clear that the two products are priced separately, and its split between AI that assists and AI that acts is coherent.
Overlap of this kind is common when a vendor launches a new product beside an established one. For a period both run, both are sold, and the customer holds entitlements in each. Suite consolidations in enterprise software have often left buyers holding capacity they paid for and no longer used, and the remedy was usually negotiated at renewal, through conversion credits or a migration path, rather than offered by default. A written conversion or retirement policy from Microsoft would settle the question.
Procurement and SOC leads should map every alert type to the agent and the meter that handles it, and size SCU commitments to what Security Copilot will still do. Before the next Security Copilot renewal, they should ask Microsoft in writing whether SCUs can be converted or applied to Perception, and whether the Security Copilot triage agent will be retired, and should hold any increase in provisioned capacity until the answer arrives.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1How are Security Compute Units sold and included?
By the hour, with an allowance. Microsoft’s Security Copilot pricing page says provisioned capacity, with a minimum of one SCU, is priced per hour and billed monthly, with overage billed for on-demand use. A worked example on the same page uses $4 per provisioned SCU and $6 per overage SCU. Its FAQ says eligible E5 and E7 customers receive 400 SCUs a month for every 1,000 user licenses, up to 10,000 a month, and that customers not yet using Security Copilot will be activated in phases. The page mentions Microsoft Perception only in its navigation, which now links to the newer Perception page.
In the source’s own words
2Which triage agent runs on which meter?
Two agents, two prerequisites. Microsoft’s Learn page for the Security Copilot Security Alert Triage Agent lists provisioned SCU capacity as a prerequisite, and tells customers to make sure they have enough SCUs for healthy agent operation. Its email triage is generally available and its cloud and identity triage are previews. The Perception Triage Agent’s page mentions capacity limits that administrators can set, and no SCU prerequisite. The Perception product page describes Security Copilot as AI that assists and Perception as AI that acts, priced separately.
What the two prerequisites tell a buyer. The two Learn pages describe agents that look alike to an analyst and differ sharply to a budget holder. One draws on SCUs the customer has provisioned, and warns against running short. The other has limits set by administrators and no SCU prerequisite. A SOC that switches on both may find its alert queue split across two meters, with no single view of what each alert cost. Until Microsoft says otherwise, the safer planning assumption is that SCUs fund the Security Copilot agent only, and that Perception’s spending comes on top.
Part 3The Money
The Sentinel Entitlement: Three Microsoft Pages, Three Versions Of The Offer
Greyhound SpotlightMicrosoft’s Sentinel blog offers eligible customers added data at $2.4 per GB. Its documentation, last updated in September, tells customers with a working Sentinel workspace not to disconnect it to qualify.
The SIEM is the SOC’s memory. It collects logs from across the company so analysts and agents can search them. Microsoft’s SIEM, Sentinel, is billed mainly by the volume of data it takes in, so SIEM cost grows with the business. At Ignite, Microsoft offered E5, E7, and Defender Suite customers the full Sentinel feature set inside Defender, longer retention for Defender data, and a lower price for adding other data. For a Microsoft-heavy company, that could reset the SIEM budget.
Cheaper data changes behavior as well as the bill. When ingestion was costly, SOCs chose which logs to keep, itself a security decision. Cheaper ingestion invites teams to bring in more, which can improve what agents and analysts see, and can also enlarge the total bill even at a lower rate. Greyhound Research expects many E5 and E7 customers to test the offer at their next SIEM renewal, and rival SIEM vendors to respond on price. A point that is easy to miss is that the value depends on the data mix. A company whose logs come mostly from Microsoft products gains most, and one with large volumes from other sources gains less than the headline suggests.
Table 3 sets the offer as three Microsoft pages state it. Retention is how long data is kept for searching. Ingestion is the act of bringing data into the SIEM.
Greyhound Research · Table 3 · The SIEM offer, page by page
Three Microsoft pages, three versions of one offer
The Sentinel SIEM offer for E5, E7, and Defender Suite customers as three Microsoft pages state it, November 19, 2026.
The direction is clear and the terms are not. Sign one version, with retention and eligibility stated, in the order form.
| Term | Microsoft Security Blog | Microsoft Sentinel Blog | Microsoft Learn |
|---|---|---|---|
| Status | Core generally availableFull SIEM breadth in preview | Not labeled Opt in from November 17 | Preview |
| Who is eligible | Defender Suite, E5, and E7 customers | Existing Sentinel customers with E5, E7, or Defender Suite | Eligible customers without an active Sentinel workspace |
| Retention of Defender data | 90 days by default | 90 days, up from 30 | 30 days included |
| Price of added data | “reduced ingestion rates” | $2.4 per GB across 500+ connectors | “Additional ingestion charges might apply” |
| Data lake | Built in | Included | Not mentioned |
| Page date | November 17, 2026 | November 17, 2026 | Updated September 23, 2026 |
Table 3 sets out the Sentinel SIEM offer for E5, E7, and Defender Suite customers as three Microsoft pages state it, and shows them differing on status, eligibility, and retention. The Microsoft Security Blog of November 17 says the core is generally available with full SIEM breadth in preview, offers 90 days of default retention, reduced ingestion rates, and a built-in data lake. The Microsoft Sentinel Blog of November 17 lets existing Sentinel customers opt in from November 17, with 90 days of retention, up from 30, added data at $2.4 per GB across 500 or more connectors, and an included data lake. Microsoft Learn, updated September 23, calls it a preview for customers without an active Sentinel workspace, with 30 days of retention and possible additional ingestion charges.
The three pages agree on the direction: a cheaper SIEM, bundled with Microsoft’s top suites. They disagree on almost every detail a buyer would sign. One says the core is generally available, one does not label it, and one calls it a preview. One says Defender data is kept for 90 days, another for 30. One welcomes existing Sentinel customers, and the oldest is written for those without a Sentinel workspace. The newest pages are probably right, but a contract needs one version.
Greyhound Standpoint:
Greyhound Research concludes that the E5 and E7 SIEM entitlement is the most consequential pricing move in Microsoft security this year, and the least clearly documented. Because SIEM cost grows with data volume, a lower rate for added data and longer included retention change the budget every year. For companies already on E5 or E7, the offer makes Microsoft the default SIEM to beat on price.
When a suite vendor folds a category into its top licence, the standalone price tends to reset, and the buyer’s question moves to fit and the cost of leaving later. Greyhound Research wrote in June 2025 that Microsoft was turning security from a revenue line into a loyalty lock. The SIEM offer is the clearest case yet, and the saving is now concrete, so the view has moved from warning to condition: take the saving, and keep the exit priced. In September, Greyhound Research also said that Microsoft’s claim of more than 500 connectors establishes reach, not equal treatment of every source. That test still applies. The documentation is a second weak point, with three Microsoft pages differing on status, retention, and eligibility.
CISOs and procurement leads should model current ingestion against the $2.4 per GB rate, reproduce their most valuable cross-vendor detections before retiring another SIEM, and get retention, eligibility, and status confirmed in the order form. Opting in is optional, so the decision belongs at the next SIEM renewal or architecture review.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What does each page say?
The security blog. Holeček writes that eligible Defender Suite, E5, and E7 customers receive all Sentinel SIEM features, 90-day default retention for Defender data, more than 500 connectors, reduced ingestion rates, and a built-in data lake, and that the complete breadth of SIEM capabilities is in preview. The Book of News says the core of the integrated SOC console is generally available.
The Sentinel blog. Microsoft’s Sentinel team wrote that from November 17, 2026, existing Sentinel customers with E5, E7, or Defender Suite can opt in, with 90 days of retention for eligible Defender data, up from 30, and added data at $2.4 per GB.
The documentation. Microsoft’s Learn page on the console, last updated September 23, says it is in preview, offers 30 days of included retention, and is available to eligible customers without an active Sentinel workspace.
What the three versions tell a buyer. The pages differ in date as well as detail. The documentation predates Ignite, and the two blogs were written for it, so the blogs most likely describe the current offer. A buyer cannot sign on likelihood, however. Retention decides how far back an investigation can look without extra cost, and eligibility decides whether a company with an existing Sentinel workspace qualifies at all. Those two terms, more than the rate for added data, decide what the offer is worth to a company that already runs Sentinel.
In the source’s own words
- Microsoft Security Blog, Holeček at IgniteThe complete breadth of SIEM capabilities is now in preview.Source
- Microsoft Sentinel Blog, November 17Lower-cost ingestion at $2.4 per GB to broaden visibility across Microsoft and non-Microsoft data through 500+ connectors.Source
- Microsoft Sentinel Blog, November 17Opting in is optional.Source
- Microsoft Learn, ISOC overviewEligible customers receive 30 days of included retention for Defender data during this phase of the preview.Source
- Microsoft Learn, ISOC overviewDon't disconnect a production Microsoft Sentinel workspace only to qualify for the ISOC preview.Source
2How does Sentinel normally charge?
By tier and volume. Microsoft’s Sentinel billing documentation says pricing depends on the tier data is ingested into. The analytics tier is billed pay-as-you-go or through commitment tiers starting at 100 GB a day, and the data lake tier is billed for ingestion, processing, storage, and queries. Security alerts from Defender products are free, while some raw Defender, Entra, and other logs are paid. The page gives no list prices and does not mention the E7 suite.
Where E7 fits. Microsoft made its Microsoft 365 E7 suite generally available on May 1, 2026, at $99 per user. It bundles E5, Microsoft 365 Copilot, and Agent 365, with the Entra Suite and advanced Defender, Intune, and Purview capabilities. The March announcement does not mention Sentinel or Security Copilot.
Part 4The SOC And Its Rivals
Staffing And Sourcing: Fewer Sorters, More Judges
Greyhound SpotlightMicrosoft sells Perception with its own Frontier Security Engineers, services teams, and partners working beside customers. The company that sells the agents also offers the people who run them.
Every SOC leader knows the shape of the problem Perception is sold to solve: more alerts than analysts can read. Microsoft’s answer is an agent it says can cover about 80% of a typical Defender queue, with security teams setting priorities and approving critical actions. Microsoft also places its own Frontier Security Engineers, services teams, and partners beside customers. The question for leaders is what their own people do next, and what changes in contracts with managed security providers, the outside firms many companies pay to run their SOC.
Greyhound Research reads the change as a shift in the work, not only in its volume. Agents take over the sorting at tier 1. They also create work that did not exist before: checking the agents’ verdicts, tuning their detections, approving their actions, and reviewing their rights. Microsoft frames this as freeing analysts for the decisions that matter. Greyhound Research reads it as moving analysts from sorting alerts to judging the agents that sort them. When the same vendor also supplies the engineers, the buyer should be clear about who checks whose work.
Workforce planning is where the change will be felt longest. Tier 1 has long been how a SOC trains its people. New analysts learn what normal looks like by sorting thousands of alerts before they move on to investigation and hunting. If agents take most of that sorting, companies will need another way to build the judgment their senior analysts rely on, such as supervised audit work or rotation through detection engineering. Greyhound Research expects managed security providers to rewrite their service levels around agent oversight rather than ticket volume, since much of what they priced was tier 1 time. The second-order effect, which gets less attention than the headcount debate, is on the talent pipeline a few years out.
Greyhound Standpoint:
Perception should change what a SOC’s people do before it changes how many there are, in Greyhound Research’s view. The first year of agent triage needs more judgment, not less, because someone has to check the agents, approve their actions, and own their mistakes. Microsoft frames the change as freeing analysts for the decisions that matter most, which is fair.
The outsourcing waves of earlier decades offer a useful caution. Companies that moved operations to a provider and then cut the in-house team often found they had kept the contract but lost the ability to judge the provider’s work. Agentic triage carries the same risk in a new form. Today’s tier 1 analysts are best placed to notice when an agent sorts alerts wrongly. When the vendor that sells the agents also supplies the engineers who help run them, as Microsoft does with its Frontier Security Engineers and partners, independent review becomes a design choice rather than an assumption. A year of audit results showing a low and stable error rate would justify a smaller team.
CIOs and CISOs should redeploy tier 1 analysts to audit, detection engineering, and response, and should not cut headcount until a year of audit results shows the agents’ error rate. At signing or renewal, contracts with managed security providers, and with Microsoft’s own services teams, should name who audits agent-closed alerts, how often, and who is liable for a closure that was wrong.
The Deeper Read · About 1 minute
Select a question to open the full answer
1What does Microsoft offer beside the agents?
People as well as software. Holeček writes that Microsoft’s Frontier Security Engineers, services teams, and partners are working side by side with customers. Microsoft’s Perception blog says security teams set priorities and approve critical actions, and that analysts review detection recommendations before they are deployed. Neither page says who audits the closures the Triage Agent makes on its own.
2How do Microsoft and its rivals describe the analyst’s new role?
Governed, not replaced. Holeček writes that agents free analysts to focus on the critical decisions that matter most, and that Microsoft’s Frontier Security Engineers, services teams, and partners work alongside customers. Splunk writes that the agentic SOC is not about replacing analysts, and describes a move to work that is led by agents and governed by people. CrowdStrike says analysts can pause or stop a running agent mid-task.
What the shared framing tells a buyer. Microsoft, Splunk, and CrowdStrike describe the analyst’s future in similar terms: governed work, with people approving or stopping agents. That agreement is useful, because a buyer is unlikely to be offered a vendor that promises to remove the SOC team. It also means the vendors have left the shape of that governed work to the customer. None of the passages says how many people the oversight needs, what skills it takes, or who is accountable when an agent errs. Those answers belong in the buyer’s operating model and in its provider contracts.
Part 4The SOC And Its Rivals
Rivals Compared: Every SOC Vendor Now Sells Agents, On A Different Meter
Greyhound SpotlightSeven vendors in this comparison sell agents for the SOC, and they count usage in at least five different ways: capacity units, unnamed consumption, tokens, findings, and credits.
Microsoft is not alone. Every major SIEM and XDR vendor, and both of its large cloud rivals, now sells agents that triage or investigate alerts. XDR, extended detection and response, is the class of product that watches devices, identities, email, and cloud together. On the agents themselves, the offers look alike. They differ in what data they see natively, how much a person approves, and above all in how the buyer pays.
Read together, the rival offers show a market that has converged on the agent and diverged on the terms. Every vendor here describes triage or investigation, and most describe some form of human approval. Where they differ is in what a buyer can verify before signing: a unit price, a cap, an approval gate, or a record in production. Google’s triage agent, for instance, has been generally available since April, months before Perception. Greyhound Research expects buyers to use that variety as leverage, asking each vendor for the terms its rivals already publish. Most coverage misses the switching question. Whatever tuning and feedback a SOC builds into one vendor’s agents is unlikely to move with it to another.
Table 4 compares the seven by product, on each vendor’s own announcements. Human control means what the vendor says about a person approving or stopping an agent.
Greyhound Research · Table 4 · SOC agents by product
Seven vendors sell SOC agents, and no two meter them alike
Agentic triage and investigation from Microsoft and six rivals, compared by product on each vendor’s own announcements, November 19, 2026.
The agents look alike and the meters do not. Compare on cost per alert handled correctly, measured on your own alerts.
| Vendor and product | Agentic triage or investigation | How agents are paid for | Human control stated |
|---|---|---|---|
| Microsoft Perception | Generally available Triage, investigation, detection, and threat intelligence agents | Consumption, no published rate. Security Copilot separately in SCUs | Approval controls for critical actions. No approval before a triage closure |
| Google Security Operations | Generally available Triage and Investigation agent since April 2026Hunting and detection agents in preview | Agentic security tokens, with budgets per agent | Not stated in the sources read |
| AWS Security Incident Response | Available Agentic investigation, no additional costSecurity Agent penetration testing generally available | Per finding, first 10,000 a month free | Agent asks the case submitter questions |
| CrowdStrike Falcon | Generally available Coordinated multi-agent investigationsAgent insider detection in the fourth quarter | Credits, with per-agent visibility and caps due in the fourth quarter | Approval gates on sensitive agent actions. Analysts can pause or stop an agent |
| Palo Alto Networks Cortex AgentiX | Released, status per agent not stated Case Investigation agent | Not stated in the sources read | Takes action with analyst oversight |
| Splunk Enterprise Security | Launched, status not stated Triage Agent and others, June 2026 | Not stated in the source read | Agent-led and human-governed |
| SentinelOne Purple AI | Open to all customers as a trial Agentic Investigation, June 2026 | Singularity Credits, not charged during the trial | Adjustable human in the loop |
Table 4 compares agentic triage and investigation from seven vendors by product, and shows that each meters its agents differently. Microsoft Perception: generally available agents, consumption pricing with no published rate, and approval controls for critical actions but none before a triage closure. Google Security Operations: a generally available Triage and Investigation agent, paid in agentic security tokens with budgets per agent. AWS Security Incident Response: agentic investigation at no additional cost, in a service priced per finding with 10,000 free a month. CrowdStrike Falcon: generally available multi-agent investigations, paid in credits with per-agent visibility due in the fourth quarter, with approval gates on sensitive actions and analysts able to pause or stop an agent. Palo Alto Networks Cortex AgentiX: a Case Investigation agent that acts with analyst oversight. Splunk Enterprise Security: a Triage Agent launched in June 2026, agent-led and human-governed. SentinelOne Purple AI: Agentic Investigation open to all customers as a trial, metered in Singularity Credits that are not charged during the trial, with adjustable human control. Where a vendor’s source does not state pricing, the table says so.
Microsoft’s advantages are breadth and bundling: agents across red, blue, and green work, inside the console that E5 and E7 customers already own, with a cheaper SIEM attached. Google has had a generally available triage agent since April. AWS gives its investigation agent away inside a service priced per finding. The others sell agents on their own platforms and credits. No two vendors meter alike. The fairest comparison is cost per alert handled correctly, which every buyer has to measure for itself.
Greyhound Standpoint:
Greyhound Research’s position is that agentic triage is becoming a standard feature of every SOC platform, and that the contest is moving to data reach, human control, and price clarity. Microsoft leads on breadth and on bundling for Microsoft-heavy estates. Every rival in Table 4 sells its own agents. Only AWS publishes a unit price, for a narrower service, and no vendor publishes an error rate. Greyhound Research said in September that Microsoft had not established that one supplier is technically necessary for an agentic SOC, and general availability does not change that.
Capabilities that begin as differentiators in security tend to become expected features within a few product cycles, as antivirus did inside endpoint protection suites and spam filtering did inside mail platforms. Buyers then choose on integration, data coverage, and running cost. The pricing units here make that comparison harder than usual. Capacity units, unnamed consumption, tokens, findings, and credits cannot be converted into one another from published material. A common unit, or published error rates from any vendor, would make the market easier to read.
SOC leads should run any comparison as a timed trial on the company’s own alerts, measuring cost per alert and the share of closures an analyst would have reversed. A rival’s agent that is cheaper per unit but wrong more often is not cheaper. The trial results belong in the platform decision at the next SIEM or XDR renewal.
The Deeper Read · About 2 minutes
Select a question to open the full answer
1What do Google and AWS offer?
Google. Google’s Triage and Investigation agent for Security Operations has been generally available since April 2026, and its new threat hunting and detection engineering agents are previews. Google says customers will be able to set token budgets per agent, and warns that every workload needs to justify its token burn.
AWS. AWS made agentic investigation in AWS Security Incident Response available to all its customers at no additional cost in November 2025, and prices the service per finding, with the first 10,000 findings a month free. Its AWS Security Agent, now part of AWS Continuum, reached general availability for on-demand penetration testing on March 31, 2026, in six Regions, and its documentation covers code review, threat modeling, and design review, not the SOC.
In the source’s own words
- Google Cloud, July 10, 2026High compute is the enemy of ROI, so every workload needs to justify its token burn.Source
- AWS What’s New, November 21, 2025Customers pay $0.000676 per finding after the free tier, with tiered discounts that reduce rates as volume increases.Source
- AWS What’s New, November 21, 2025automatically enabled for all Security Incident Response customers at no additional costSource
2What do the SIEM and XDR vendors offer?
Four platforms. CrowdStrike made coordinated multi-agent investigations generally available and set agent insider threat detection for the fourth quarter. It lists human-in-the-loop gates, under which an agent pauses for an analyst’s approval before a sensitive tool call, as available today, and credit visibility and caps for each SOC agent for the fourth quarter. Palo Alto Networks says its Cortex AgentiX Case Investigation agent takes action with analyst oversight. Its frontier model support was a private preview in June, and its July 2026 release says SOC teams can now put frontier models to work. Splunk launched a Triage Agent among others in June 2026, without a stated status. SentinelOne opened Purple AI Agentic Investigation to all customers in June as a trial, metered in Singularity Credits that are not charged during the trial, with autonomy the customer can adjust.
Before, after, or not at all. The vendors describe human control in different ways, and the differences are worth reading closely. CrowdStrike pauses an agent before a sensitive tool call. SentinelOne lets the customer adjust autonomy. Palo Alto Networks speaks of analyst oversight. Microsoft’s Triage Agent, by its documentation, closes false alarms without an approval step. No single model suits every SOC, but together they give a buyer a concrete test for any trial: for each action an agent can take, does a person approve it before, review it after, or not see it at all?
In the source’s own words
- CrowdStrike, November 4, 2026When an agent invokes a sensitive tool, it pauses and raises an Action Request.Source
- Palo Alto Networks blog, February 25, 2026takes action with analyst oversight.Source
- Palo Alto Networks blog, July 30, 2026SOC teams can now put leading frontier AI models to work across security operationsSource
- SentinelOne, June 17, 2026Customers set the degree of autonomy through an adjustable human-in-the-loop approach.Source
Part 5The Risks And The Counter-Argument
Risks To Watch: Where An Agentic SOC Can Go Wrong
Greyhound SpotlightNone of the five risks needs an attacker to defeat Perception. Each comes from trusting it without checking.
Greyhound Research sees five risks in moving the SOC’s first line to agents. Each starts inside the buyer’s own organization, and each shows an early warning in a report, a contract, or a change log the buyer already holds.
None of these is the kind of risk security teams usually plan for. Security registers have tended to list threats from outside, such as an intruder, a malware family, or a supplier breach, each with a control that blocks it. The five here come from inside the control system itself, from a defender that is trusted too quickly or given too much room. That makes them hard to see on the usual dashboards, which report what was caught rather than what was closed in error. Greyhound Research expects auditors and regulators to ask about the oversight of security agents before long, as they have for other automated decisions. Preparing that evidence now costs less than assembling it against a deadline.
Table 5 sets each risk beside its early warning and its owner. Every early warning can be checked without help from Microsoft.
Greyhound Research · Table 5 · Risk register
Every risk comes from trusting the agents without checking
Five risks in moving the SOC’s first line to agents, each with its early warning and its owner.
Each risk is cheaper to catch than to unwind, and each early warning can be checked this quarter without Microsoft.
| Risk | Early warning in the buyer’s own estate | Owner |
|---|---|---|
| Closures left uncheckedThe Triage Agent closes a real attack as a false alarm, and no one re-examines it. | A closure report with no sampled re-examination. | SOC lead |
| Defenders with broad rightsA security agent holds more scope than its task, or anyone can start it. | A role list in which every Security Reader can use every agent. | Chief information security officer |
| Shields left in placeA temporary network control outlives its purpose, or ends before the patch. | A change log with shields that carry no expiry or patch owner. | Head of IT operations |
| A meter without a ceilingAgent consumption rises with alert volume and draws on the Azure commitment. | An invoice line for Perception with no cap or budget owner. | Head of procurement, with the CFO |
| Terms with no single versionSCUs go unused, or the SIEM offer is read differently by each party. | An order form that cites a page still saying preview. | Head of procurement |
Table 5 lists five risks in moving the SOC’s first line to agents, each with an early warning in the buyer’s own estate and an owner. Closures left unchecked, warned by a closure report with no sampled re-examination, owned by the SOC lead. Defenders with broad rights, warned by a role list in which every Security Reader can use every agent, owned by the CISO. Shields left in place, warned by a change log with shields that carry no expiry or patch owner, owned by the head of IT operations. A meter without a ceiling, warned by an invoice line with no cap or budget owner, owned by procurement with the CFO. Terms with no single version, warned by an order form citing a page that still says preview, owned by procurement. The risks are Greyhound Research analysis.
The five share a cause: the belief that a system which acts on its own also checks itself. Each warning sits in something the buyer already produces, such as a closure report, a role list, a change log, an invoice, or an order form. The table names owners, not likelihoods, and ranks none of the risks above another. This is a judgment about order, not a prediction of harm.
Greyhound Standpoint:
In Greyhound Research’s reading, none of the five risks is new. Each is an old control applied to a new kind of worker. Sampling decisions, limiting privilege, expiring temporary changes, capping spend, and fixing contract terms are disciplines most enterprises already run for staff, service providers, and software. What is new is that this worker acts at machine speed and around the clock, so a gap in any of these controls compounds faster than it would with a human team.
The history of internal control supports a calm reading. Segregation of duties, change management, and the four-eyes principle each grew from the same observation: a capable actor working alone will in time make an error that goes unseen until it is costly. None of these controls assumes bad intent, and none of them stops the work. They make errors visible early, while they are cheap to correct. Unchecked closures, broad rights, forgotten firewall changes, open-ended spending, and unclear contracts are all cheaper to prevent than to unwind, and Microsoft supplies part of the tooling. The register ranks none of the risks above another, because the evidence does not yet support a ranking. A year of the buyer’s own audit data would.
Buyers can check each early warning this quarter, in reports they already produce, and each risk has an owner in the Runbook that follows. The register itself belongs on the agenda of the next board risk review.
Part 5The Risks And The Counter-Argument
The Strongest Objection: The Alert Queue Cannot Wait For An Audit
Greyhound SpotlightThe strongest objection to this note is that SOCs cannot afford to slow down the one tool that clears their backlog. That part stands.
The objection deserves its full weight. SOC teams face more alerts than they can read. Most alerts are harmless, and an agent that clears them frees analysts for real attacks. Microsoft has built approval controls, audit logs, and least-privilege access into Perception. For companies already on E5 or E7, the cheaper SIEM makes the whole package attractive. On this view, adding an audit adds cost to the very work the agent was meant to save, and can wait until the agent has proved itself.
The objection draws its strength from a real constraint. Most SOCs are short of staff, and leaders who have carried an alert backlog for years will reasonably resist anything that slows the tool clearing it. It also reflects a sound instinct about new tools, which is to prove their value before adding overhead. What it leaves open is who would supply that proof. Microsoft publishes coverage, not accuracy, and no rival in this note publishes an error rate either. The only party holding both the data and the reason to measure an agent’s mistakes is the customer. Greyhound Research expects that to remain the case for at least the first year of general availability.
Exhibit 6 tests each part of the objection against the evidence. Three parts stand. The claim that the audit can wait does not hold, because the agent closes alerts from its first day.
Greyhound Research · Exhibit 6 · The objection held to the evidence
Three parts of the objection stand, and one falls
The fair counter-case to this research note, part by part, against the evidence gathered for each.
The case for the agents stands, and the case for a later audit does not. Use them now, and sample their closures from the first day.
Exhibit 6 holds four parts of the strongest objection to this research note against the evidence, as four cards, and shows that three stand and one falls on timing. SOCs cannot keep up with alerts, the problem Microsoft built the Triage Agent to solve: stands. The bundle lowers SIEM cost for E5 and E7 customers: stands. Microsoft has built least-privilege, approval, audit, and usage controls: stands. The audit can wait: falls on timing, because the agent closes alerts from its first day and only an audit can show it deserves more trust. The verdicts are Greyhound Research analysis.
Greyhound Standpoint:
Greyhound Research accepts the objection on the backlog, on the bundle, and on Microsoft’s controls, and rejects it on timing. The case for agents in the SOC is strong. Alert volumes have outgrown analyst time, the E5 and E7 SIEM offer lowers the cost of the data behind the agents, and Microsoft has built approval controls and audit logs into the product. Buyers should use the agents.
Timing is where the objection does not hold. The Triage Agent closes alerts from its first day, so the question of whether it can be trusted is being answered in production whether or not anyone measures it. Without a sample of closures checked from the start, there is no evidence on which to trust the agent more later, and the first sign of an error may be an attacker who returns. The logic is that of any new hire. A capable new analyst is given real work in the first week and has that work reviewed during a probation period, with review lightened as the record builds. The agent deserves the same treatment. The audit is how the agent proves itself, not a tax on its savings.
A weekly sample takes a few analyst hours and does not slow the queue. SOC leads should start the sample on the day the agent is switched on, and CISOs should set, at the pilot gate, the error rate at which review can be reduced.
The Deeper Read · About 1 minute
Select a question to open the full answer
1Why does the timing claim not hold?
Closures start on day one. Microsoft’s Triage Agent documentation says the agent runs automatically once set up and resolves false alarms itself. Microsoft’s own figure for coverage is not a guaranteed rate for any one customer. A customer that waits for the agent to prove itself, without measuring it, will have no proof to wait for. Greyhound Research’s November 15 note on the Microsoft 365 outage recalls the firm’s public statement of November 13 that enterprises “have priced Microsoft 365 as a licence and never as a single point of failure.” A SOC that relies on one vendor’s agents without checking them would repeat that pattern.
Why the outage comparison fits. The comparison is not a forecast that Perception will go wrong as the suite did. It is that both rest on one vendor’s system and on an assumption no one has tested. In the outage, the assumption was that the suite would stay up. In the SOC, it is that the agent’s closures are right. A weekly sample of closures is the SOC’s equivalent of a continuity test: modest in cost, run before it is needed, and the only way to learn how much reliance is safe.
In the source’s own words
- Greyhound Research, Microsoft 365 Outage note, recalling its November 13 statementEnterprises have priced Microsoft 365 as a licence and never as a single point of failure.Source
Part 6What To Do, By Role
The Greyhound CXO Runbook: Ten Moves Before The Agents Run Unwatched
Greyhound SpotlightEvery move uses something the buyer already controls: a closure report, a role list, a change process, a pilot gate, an order form, an invoice, a business case, a staffing plan, or the board’s risk agenda.
Ten moves follow from the evidence, at least one for each leader who will own part of an agentic SOC. Each names the assumption it corrects, the evidence behind it, and the executive who decides. None waits for Microsoft, because each uses controls the buyer already holds.
The moves are deliberately ordinary. Each extends a control that most enterprises already run for people and suppliers: sampling, access review, change approval, production gates, purchase approval, contract terms, business cases, staffing plans, and board oversight. That choice reflects how new technology tends to be absorbed in practice. Programs built from scratch for a new category often stall, while extensions of an existing discipline quickly gain owners, budgets, and audit trails. Greyhound Research expects the first two moves, the weekly sample and the rights review, to return the most in the first quarter, because they produce evidence the other eight can draw on. A sample that shows a low error rate also makes the later conversations with finance and the board easier.
Table 6 sets the ten moves beside the assumption each corrects, its evidence, and its owner. Two of the ten finish only when Microsoft publishes its terms.
Greyhound Research · Table 6 · The Greyhound CXO Runbook by owner
Ten moves, and none waits on Microsoft to start
Ten moves for the leaders who will own part of an agentic SOC, with the assumption each corrects, its evidence, and its owner.
Start all ten this quarter. Only the rate card and the SCU and SIEM terms wait on Microsoft to finish.
Select a move to open its full text in The Deeper Read below
| Move | The assumption it corrects | Evidence anchor | The decision sits with |
|---|---|---|---|
| 1/Audit a sample of agent-closed alerts every week | Coverage is accuracy | Exhibit 2 | SOC lead |
| 2/Govern every security agent as a privileged account | The defenders’ agents are not insiders | Exhibit 3 | CISO |
| 3/Put an expiry date and a patch owner on every shield | A shield is a fix | Exhibit 4 | Head of IT operations, with the CISO |
| 4/Get a rate card and a cap before any commitment | An estimator is a price | Exhibit 5 | Head of procurement, with the CFO |
| 5/Reconcile SCUs and the SIEM offer in one order form | Each security product renews on its own | Tables 2 and 3 | Head of procurement, with the SOC lead |
| 6/Move people from sorting to judging before cutting any | Agents save headcount in year one | Section 4.1, Table 4 | CIO, with the CISO |
| 7/Admit only generally available agents and connections to production | General availability covers everything announced with it | Table 1 | CIO |
| 8/Put every playbook and agent setting under change control | Agent setup is a one-time product setting | Exhibit 2, Section 1.1 | Head of IT operations |
| 9/Approve the business case on cost per alert handled correctly | The lowest unit price is the lowest cost | Table 4, Exhibit 5 | CFO |
| 10/Review the audit of the agents every quarter | Agent oversight is an operating matter, not a board one | Table 5, Exhibit 7 | Board risk committee |
Table 6 sets out the ten moves of the Greyhound CXO Runbook, each with the assumption it corrects, the exhibit or table it rests on, and the executive with whom the decision sits, and shows that none waits on Microsoft to start while two complete only when Microsoft answers. The moves are: audit a sample of agent-closed alerts every week, govern every security agent as a privileged account, put an expiry date and a patch owner on every shield, get a rate card and a cap before any commitment, reconcile SCUs and the SIEM offer in one order form, move people from sorting to judging before cutting any, admit only generally available agents and connections to production, put every playbook and agent setting under change control, approve the business case on cost per alert handled correctly, and review the audit of the agents every quarter. The moves are Greyhound Research analysis.
The two that finish later, on the rate card and on the SCU and SIEM terms, wait on answers Microsoft has not yet published. Those answers are what the questions that follow ask for. Every other move rests on an exhibit or table in this note and names who decides. The remaining eight can be finished with controls the buyer already holds.
The Deeper Read · About 5 minutes
Select a move to open its full text
1SOC leads: audit a sample of agent-closed alerts every week.
Exhibit 2 shows that the Triage Agent closes false alarms with no approval step. This move corrects the belief that coverage is accuracy. Its action is to draw a random sample of agent-closed alerts each week, have an analyst re-examine them, and report the share that should have stayed open, by alert type, to the CISO. The decision sits with the SOC lead.
2CISOs: govern every security agent as a privileged account.
Exhibit 3 shows Perception’s agents holding power over alerts, detections, and live systems. This move corrects the assumption that the defenders’ agents are not insiders. Its action is to give each agent an owner, a written list of actions it may take alone, approval before changes to live systems, and a quarterly rights review, and to limit who may start an agent. The decision sits with the CISO.
3Heads of IT operations: put an expiry date and a patch owner on every shield.
Exhibit 4 shows the shield written by Microsoft and the patch left to the customer. This move corrects the belief that a shield is a fix. Its action is to agree who approves a shield on a critical system, set a maximum life for each shield, and name the patch owner in the change log before shielding is turned on. The decision sits with the head of IT operations, together with the CISO.
4Procurement: get a rate card and a cap before any commitment.
Exhibit 5 shows a meter with no published rates. This move corrects the assumption that an estimator is a price. Its action is to obtain unit rates in writing, a cap per agent and per month, and a rule on who may approve Perception drawing on an Azure commitment, before signing an AI Flex Plan or counting Perception against a MACC. The decision sits with the head of procurement, together with the CFO.
5Procurement and SOC leads: reconcile SCUs and the SIEM offer in one order form.
Tables 2 and 3 show two meters for one alert queue and three versions of the SIEM offer. This move corrects the habit of renewing each security product on its own. Its action is to map each alert type to the agent and meter that handles it, size SCU commitments to that map, and get the retention period, eligibility, and status of the SIEM offer confirmed in the same order form. The decision sits with the head of procurement, together with the SOC lead.
6CIOs: move people from sorting to judging before cutting any.
Section 4.1 and Table 4 show the work shifting from sorting to judging, and agents on every platform. This move corrects the belief that agents are a headcount saving in their first year. Its action is to redeploy tier 1 analysts to audit, detection, and response, hold headcount until a year of audit results shows the agents’ error rate, and write the audit into managed security contracts. The decision sits with the CIO, together with the CISO.
7CIOs: admit only generally available agents and connections to production.
Table 1 shows Perception, seven of its agents, and the core of the integrated SOC console as generally available, while the AWS, Okta, and Salesforce signals, vulnerability shielding, the Web App Pen Test and Risk Reduction agents, and Dynamic Agent Risk remain previews. This move corrects the assumption that general availability covers everything announced with it. Its action is to use Table 1 as the checklist at the pilot gate it already runs. Only the generally available rows go into production, and each preview row stays in a pilot schedule with its own exit terms. For every agent switched on, someone checks that the launch post and the agent’s Learn page agree on what it does, and repeats the check whenever a page changes, since three Microsoft pages still describe Perception as a preview. The list in the order form stays with procurement. The decision sits with the CIO, at the pilot gate and again at the first renewal.
8Heads of IT operations: put every playbook and agent setting under change control.
Exhibit 2 and Section 1.1 show agents that run automatically once set up, a setup that switches off the built-in tuning rules the agent replaces, and playbooks that compose teams of agents and set them to act on their own. This move corrects the assumption that agent setup is a one-time product setting. A change to a playbook can change what the SOC does without any agent changing. Its action is to treat each playbook, each agent’s alert scope, and its capacity and data access limits as change-controlled configuration, with an owner, a version history, and a review before each edit goes live. The review should also record the tuning rules the setup switched off and name a person who sees the whole loop when one agent writes detections and another closes the alerts they raise. The decision sits with the head of IT operations, at the next architecture review, and later edits go to the change advisory board.
9CFOs: approve the business case on cost per alert handled correctly.
Table 4 shows seven vendors that count usage in at least five ways, and capacity units, unnamed consumption, tokens, findings, and credits cannot be converted into one another from published material. This move corrects the assumption that the lowest unit price is the lowest cost. A rival’s agent that is cheaper per unit but wrong more often is not cheaper. Its action is to approve no business case for SOC agents, through the approval process the CFO already runs, until it states the cost per alert handled correctly. That figure is a month of actual spend from the invoice, divided by the alerts the agents handled, less the share that the weekly sample in the first move shows should have stayed open. The same month should test the cost estimator in Exhibit 5 against actual usage before any figure from it enters a forecast. The decision sits with the CFO, at the next SIEM or XDR renewal.
10Board risk committees: review the audit of the agents every quarter.
Table 5 sets five risks against early warnings that sit in reports the buyer already produces, and Exhibit 7 names four signals, all controlled by Microsoft, that would support lighter supervision. This move corrects the assumption that agent oversight is an operating matter that dashboards already cover. Those dashboards report what was caught, not what was closed in error. Its action is to put the risk register on the next board risk review agenda and ask the CISO each quarter for the five early warnings: the sampled share of closures that should have stayed open, the result of the rights review, open shields without an expiry or patch owner, spending against its cap, and the version of the terms cited in the order form. Once a year the committee should note which of the four signals has arrived, as context for any proposal to reduce review. The decision sits with the board risk committee.
Part 6What To Do, By Role
Questions For Microsoft: What Buyers Will Ask Next
Greyhound SpotlightFour of the six questions have no answer at all in what Microsoft has published, including the price of a unit of Perception.
Every finding in this note rests on what Microsoft has published, and Microsoft alone can close the gaps it names. Six questions follow, in the order the evidence raises them. Each gives what Microsoft has said so far, what the answer would settle, and the Runbook move it maps to. No move waits on them, and an account team can answer each from material Microsoft already holds.
Each question asks for something Microsoft already measures or decides: how often its agents are overturned, what rights they start with, what a unit costs, and which page governs. That makes the questions fair to put to an account team, and fair to put in writing. Greyhound Research expects the commercial questions, on the rate card, SCUs, and the SIEM offer, to be answered soonest, because clear terms help Microsoft close deals. The accuracy question is likely to take longest. An error rate depends on data across many customers, varies by alert type, and invites comparison with rivals. Buyers who keep their own closure samples will have a working answer of their own well before Microsoft publishes one.
Table 7 tracks the six questions against what Microsoft has said so far. Two are answered in part, and four not at all.
Greyhound Research · Table 7 · Questions for Microsoft, tracked
Two questions answered in part, four not at all
The six questions only Microsoft can answer, against what it had stated by 01:15 UTC on November 19, 2026.
Four of six are unanswered. Ask in writing, add each answer to the order form, and act on the Runbook meanwhile.
Select a question to open its full text in The Deeper Read below
| Question | What Microsoft has said so far | The Runbook move it settles |
|---|---|---|
| Q1What share of agent-closed alerts are later found to be real, by alert type? | Not stated A coverage figure of about 80%, not guaranteed for any one customer. No accuracy figure. | Runbook move 1 |
| Q2What rights does each Perception agent hold by default, and who may start it? | Stated in part Agent identities get every scope they need, and any Security Reader can use the agents. | Runbook move 2 |
| Q3How long does a vulnerability shield last, who approves it, and how is it removed? | Not stated Shields are called temporary and can change network accessibility. | Runbook move 3 |
| Q4What is a unit of Perception, what does it cost, and can it be capped per agent? | Not stated Consumption pricing, an estimator, and a calculator. No rate card. | Runbook move 4 |
| Q5Can SCUs pay for Perception, and will the Security Copilot triage agent be retired? | Not stated The two products are priced separately. | Runbook move 5 |
| Q6Which version of the E5 and E7 SIEM offer is authoritative? | Stated in part The November pages agree on 90 days. The documentation says 30 days and preview. | Runbook move 5 |
Table 7 tracks six questions only Microsoft can answer against what it had stated by 01:15 UTC on November 19, 2026, and shows two answered in part and four not stated at all. Answered in part: the default rights of each agent and who may start it, and which version of the E5 and E7 SIEM offer is authoritative. Not stated: the share of agent-closed alerts later found to be real, how long a shield lasts and who approves and removes it, what a unit of Perception costs and whether it can be capped, and whether SCUs can pay for Perception. Each question names the Runbook move it settles. The grading is Greyhound Research analysis.
Two answers would change this note’s reading most. One is a published error rate for agent-closed alerts, by alert type. The other is a rate card with caps per agent. Either would move Perception from a strong product to a product a CFO can sign without a long negotiation. None of the six is a trick question. Each asks Microsoft to write down what its product does or costs.
The Deeper Read · About 2 minutes
Select a question to open its full text
Q1What share of agent-closed alerts are later found to be real, by alert type?
Said so far. Not stated. Microsoft gives a coverage figure of about 80% of a typical queue, which is not a guaranteed rate for any one customer, and no accuracy figure. The answer settles. The size of the weekly sample in the first Runbook move. Evidence that would settle it. Published error rates by alert type, from customer data. Greyhound Research would update. Exhibit 2.
Why accuracy comes first. Of the six, this question carries the most weight, because its answer sets how much of everything else a buyer must check. A published error rate by alert type would let a SOC sample less where the agent is reliable and more where it is not. Without one, the sample has to stay broad. The question also tests how the product is framed. Coverage describes what the agent can reach, and accuracy describes what it gets right, and buyers will want both in the same report.
Q2What rights does each Perception agent hold by default, and who may start it?
Said so far. In part. The agent identity must be granted every scope it needs, and any Security Reader can use all configured agents. Least-privilege access and approval controls exist. The answer settles. How narrow the rights in the second Runbook move can be. Evidence that would settle it. A default permission list for each agent, and a setting to limit who may start it. Greyhound Research would update. Exhibit 3.
Q3How long does a vulnerability shield last, who approves it, and how is it removed?
Said so far. Not stated. Microsoft calls shields temporary and says they can change network accessibility. The answer settles. The expiry rule in the third Runbook move. Evidence that would settle it. Documentation of shield duration, approval, and reversal. Greyhound Research would update. Exhibit 4.
Q4What is a unit of Perception, what does it cost, and can it be capped per agent?
Said so far. Not stated. Perception is sold by consumption, through pay-as-you-go or an AI Flex Plan, with an estimator and a calculator. The answer settles. The rate card and cap in the fourth Runbook move. Evidence that would settle it. A published rate card with caps. Greyhound Research would update. Exhibit 5.
Q5Can Security Copilot SCUs, bought or included, pay for Perception, and will the Security Copilot triage agent be retired?
Said so far. Not stated. The two products are priced separately. The answer settles. The SCU sizing in the fifth Runbook move. Evidence that would settle it. A written conversion or retirement policy. Greyhound Research would update. Table 2.
Q6Which version of the E5 and E7 SIEM offer is authoritative?
Said so far. In part. Microsoft’s two Ignite posts agree on 90 days of retention, and the Sentinel blog states a $2.4 per GB rate and welcomes existing Sentinel customers. The documentation still says preview, 30 days, and no existing workspace. The answer settles. The order form terms in the fifth Runbook move. Evidence that would settle it. Updated documentation and licensing terms. Greyhound Research would update. Table 3.
Part 7The Final Greyhound Standpoint
Microsoft Perception: The Final Greyhound Standpoint
Greyhound SpotlightFour signals within twelve months would move this position, and Microsoft controls all four.
Greyhound Research’s final judgment is that Ignite 2026 gave the SOC a workforce of agents, and that Microsoft has shipped it faster and more broadly than most of its rivals. That is a real achievement, and buyers should put it to work. It also changes where control sits. When agents close the alerts, the audit becomes the control: a regular, independent check of what the agents decided, run by people the buyer trusts.
The reason is that Perception’s three most consequential features each move a decision from a person to software. The Triage Agent closes alerts without asking. The shield changes live systems. The price runs on a meter that follows alert volume. Each is sound in design, and each removes a moment at which someone used to look. The audit puts that moment back, at a cost measured in analyst hours rather than headcount.
Buyers with long memories have seen this shape before. Outsourcing, the move to cloud, and successive rounds of suite consolidation each delivered their promised savings to the organizations that kept the ability to check the work, and disappointed many of those that handed over the checking along with the work. Agentic security is likely to follow the same course. The technology is not the main variable. The discipline around it is.
On product, Microsoft leads on breadth, with red, blue, and green agents in one console, and on bundling, with a cheaper SIEM for E5 and E7 customers. Google has had a generally available triage agent longer. AWS gives investigation away inside a per-finding service. On price clarity, only AWS publishes a unit rate, and for a narrower service. The defenders’ agents are insiders, as Greyhound Research’s companion note on Azure agent infrastructure holds for every agent, and need the same narrow rights.
The evidence has limits, and they are worth stating plainly. Three Microsoft pages still call Perception a preview, and the documentation of the SIEM offer is out of date. No unit rate has been published. The 80% figure is a coverage claim from Microsoft’s own analysis and is not used as a measure of accuracy. Statuses and claims are as each company published them through 1:15 a.m. UTC on November 19, 2026. Much of what this note asks of Microsoft is documentation it is well placed to supply.
Four signals would move this position within twelve months. Each would strengthen the case for letting the agents run with lighter supervision. They are a published error rate for agent-closed alerts, a rate card with caps per agent, a written path for Security Copilot SCUs, and one authoritative page for the SIEM offer.
Exhibit 7 sets those signals side by side with what each would change. All four sit with Microsoft.
Greyhound Research · Exhibit 7 · Signals that would move the position
Four signals in twelve months would move this position
What would move Greyhound Research’s judgment within twelve months, and what each would change.
Published error rates would move this view furthest, because they decide how much of the audit can safely be reduced.
Exhibit 7 lists the four signals that would move Greyhound Research’s position within twelve months, each controlled by Microsoft, and one event outside its control. Published error rates for agent-closed alerts would let buyers size the audit, the question that matters most. A rate card with caps would let finance compare agents with people and rivals. A written path for Security Copilot capacity would end the risk of paying twice. One current page for the SIEM offer would let the order form cite one version. A rival publishing rate cards and error rates first could reset the comparison. The signals are Greyhound Research analysis.
One event outside Microsoft’s control could reset the comparison: a rival publishing a full rate card and error rates for its own SOC agents first. Of the four signals, published error rates would matter most, because they decide how much of the audit a buyer can safely reduce.
Until the signals arrive, the actions are clear and sit with the buyer. SOC leads should start the weekly sample on the day the agents go live. CISOs should take the audit design and the agents’ rights to the next board risk review. Procurement should hold any AI Flex Plan, MACC drawdown, or SCU renewal until the rates and terms are in writing. Companies that do this can let the agents work the queue, check their closures every week, and sign only what is written down.
Part 8Appendix, Sources, And Disclosures
Appendix: The Evidence Register, Source By Source
Greyhound Research sets out here the evidence behind every fact, figure, and quotation in this research note, one drop-down for each class of source, so that any sentence can be checked against the words it rests on without leaving the page. Each row gives the source, when it was read, the part of the source the words sit in, what it says, and where in this research note that material is used. Times of reading are in UTC. Every source was re-read on November 19, 2026, between 00:28 and 01:15 UTC. Where a passage was summarized and not quoted, the row says so. A passage quoted in The Deeper Read of a section is not repeated here, and the source’s entry names the section that quotes it.
Microsoft announcements and blogs9 sources, 36 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Althoff, J. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in The Sentinel Entitlement. Exhibit 1Quoted in The Deeper Read of The Sentinel Entitlement | Post | [quoted in The Deeper Read] |
| Post | E7 unifies Microsoft 365 E5, Microsoft 365 Copilot, and Agent 365, with the Entra Suite and advanced Defender, Intune, and Purview capabilities (summarized, not quoted). | |
| Reading note | The post does not mention Sentinel or Security Copilot. | |
| Gallot, H. (2026a)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Ignite Changed. Exhibit 1Quoted in The Deeper Read of What Ignite Changed | Post | [quoted in The Deeper Read] |
| Post | Project Perception coordinates red, blue, and green teams of agents (summarized, not quoted). | |
| Reading note | The post gives no pricing, Security Copilot, or SCU information. | |
| Gallot, H. (2026b)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Agents As Insiders. Exhibit 1Quoted in The Deeper Read of Agents As Insiders | Post | "They are agentic insiders." |
| Post | [quoted in The Deeper Read] | |
| Holeček, A. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Microsoft Announced, Agents As Insiders, Vulnerability Shielding, Pricing, The Sentinel Entitlement, Staffing And Sourcing. Tables 1, 3, and 4. Exhibits 3, 4, and 5Quoted in The Deeper Read of Agents As Insiders, The Sentinel Entitlement, Staffing And Sourcing | Post | [quoted in The Deeper Read] |
| Post | "Eligible Defender Suite, Microsoft 365 E5, and Microsoft 365 E7 customers receive all" Microsoft Sentinel SIEM features, with "90-day default data retention for Defender data" and "more than 500 connectors, reduced ingestion rates, and a built-in data lake" (fragments). | |
| Post | [quoted in The Deeper Read] | |
| Post | "Dynamic Agent Risk assesses each agent's exposure, blast radius, and runtime behavior." | |
| Post | [quoted in The Deeper Read] | |
| Reading note | The post gives no price, unit, rate, or Security Copilot or SCU information. It does not give the 80% triage figure. | |
| Lefferts, R. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Ignite Changed. Exhibit 1Quoted in The Deeper Read of What Ignite Changed | Post | [quoted in The Deeper Read] |
| Reading note | The post gives no E5 or E7 entitlement, retention, or ingestion pricing. | |
| Microsoft (2026a)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Ignite Changed, What Microsoft Announced, The Triage Agent, Agents As Insiders, Vulnerability Shielding, Pricing, The Sentinel Entitlement. Tables 1 and 3. Exhibits 3 and 4Quoted in The Deeper Read of What Ignite Changed, The Triage Agent, Agents As Insiders, Vulnerability Shielding | Perception | [quoted in The Deeper Read] |
| Perception | [quoted in The Deeper Read] | |
| Perception | "The Triage Agent now handles AWS and Salesforce alerts and incidents." | |
| Perception | "The Investigation Agent can reason over AWS and Okta data." | |
| Perception | [quoted in The Deeper Read] | |
| Defender | "Core capabilities in the Integrated Security Operations Center (ISOC) are generally available." | |
| Defender | [quoted in The Deeper Read] | |
| AI Readiness Initiatives | "New AI Readiness Initiatives in Perception, in preview in January, will help organizations track progress" | |
| Reading note | The Book of News does not mention Security Copilot, Security Compute Units, E5, or E7 in its security sections, as read. | |
| Microsoft (2026b)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in The Triage Agent, Pricing, Staffing And Sourcing, The Strongest Objection. Table 1. Exhibit 6 | Post | "Microsoft Perception is now generally available." |
| Post | The 80% figure rests on Microsoft's analysis of about 8 million alerts (summarized, not quoted). | |
| Post | A table lists Okta and Salesforce as triage sources, Okta and AWS for investigation, and Okta, Salesforce, and AWS for detection authoring (summarized, not quoted). | |
| Post | Analysts review and adjust detection recommendations before deployment (summarized, not quoted). | |
| Reading note | Author shown as a Microsoft staff account. The post gives no pricing and does not mention Security Copilot or SCUs. | |
| Microsoft (2026c)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in The Sentinel Entitlement. Tables 3 and 7. Exhibit 6 | Post | "90 days of retention up from 30 days for eligible Defender data." |
| Post | From November 17, 2026, existing Sentinel customers with Microsoft 365 E5, Microsoft 365 E7, or Defender Suite can opt in (summarized, not quoted). | |
| Post | A built-in data lake is included for historical hunting, investigations, and compliance (summarized, not quoted). | |
| Reading note | Authors Dana Bargury and Vivek Kokkengada. The byline shows November 17, the page metadata November 16. The post links a transition resource naming March 31, 2027, for the Azure portal. | |
| Nadella, S. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Ignite Changed, The Triage Agent. Exhibit 1 | Post | "And I am very excited to share that Perception is now generally available." |
| Post | "Essentially, think about it like this: red team agents simulate attacks, blue team agents detect attacks" |
Microsoft product pages and documentation9 sources, 34 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Microsoft (n.d.a)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Microsoft Announced, Pricing, Security Copilot Commitments. Table 2. Exhibit 5 | FAQ | "Pay-as-you-go or via the AI Flex plan." |
| FAQ | Security Copilot and Perception are described as complementary and "priced separately" (fragment). | |
| FAQ | The AI Flex Plan is offered on a one- or three-year term (summarized, not quoted). | |
| Reading note | The page states no availability status, unit, or rate, and does not mention SCUs, E5, or E7. | |
| Microsoft (n.d.b)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Microsoft Announced, Pricing. Exhibit 5 | FAQ | "Project Perception uses consumption-based, pay-as-you-go pricing. You pay only for what you use" |
| Hero | Describes a workforce of specialized AI agents now in preview (summarized, not quoted). | |
| Reading note | Titled Project Perception. On November 18, 2026, the Security Copilot pricing page linked to it. On November 19, that link pointed to the newer Microsoft Perception page. | |
| Microsoft (n.d.c)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Security Copilot Commitments. Table 2 | FAQ | "400 Security Compute Units (SCUs) per month for every 1000 user licenses, up to 10,000 SCUs per month." |
| FAQ | "Overage capacity is billed for on-demand usage." | |
| FAQ | "so you'll be charged 4 SCUs at $4 per provisioned SCU, totaling $16 for that hour." | |
| FAQ | "plus 3.2 SCUs (at $6 per overage SCU), totaling $35.20 for that hour." | |
| Reading note | The page names Microsoft Perception only in its navigation, linked to the newer product page, and describes no transition from Security Copilot. | |
| Microsoft (2026d)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Microsoft Announced, Agents As Insiders. Exhibit 3. Table 7 | Roles | Security Admins install, configure, and remove agents, and Security Readers use them (summarized, not quoted). |
| Reading note | Last updated September 6, 2026. The page gives no licensing, billing, or region information. | |
| Microsoft (2026e)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Microsoft Announced, The Triage Agent, Agents As Insiders, Security Copilot Commitments, The Strongest Objection. Exhibits 2 and 3Quoted in The Deeper Read of What Microsoft Announced | Status | [quoted in The Deeper Read] |
| Behavior | "The agent runs automatically when a new alert is detected." | |
| Feedback | "For supported alert types, the agent can incorporate analyst feedback when explicitly provided and approved to tune its verdict analysis." | |
| Permissions | "To follow the principle of least privilege, assign the agent identity only the permissions the Triage Agent needs to perform its tasks." | |
| Learning | "This capability is currently available for email and collaboration alerts only." | |
| Permissions | Each alert type needs Security Copilot (read), Security data basics (read), and Alerts (manage), with email content rights added for email and collaboration alerts (summarized, not quoted). | |
| Reading note | Last updated August 19, 2026. The page describes no approval step before a closure, gives no accuracy figure, and does not mention AWS, Okta, or Salesforce. | |
| Microsoft (2026f)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in The Triage Agent, Security Copilot Commitments. Table 2 | Prerequisites | Provisioned capacity in Security Compute Units is a prerequisite (summarized, not quoted). |
| Status | "Email and collaboration alert triage capabilities are already generally available (GA)." | |
| Reading note | Last updated July 2, 2026, with metadata of August 11. The page does not mention Perception or any retirement, and gives no accuracy figures. | |
| Microsoft (2026g)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in The Sentinel Entitlement. Table 3 | Status | "ISOC is in preview. Capabilities and availability might change during the preview period." |
| Eligibility | Available to eligible Defender Suite, E5, or E7 customers that do not have an active Microsoft Sentinel workspace (summarized, not quoted). | |
| Data | "Additional ingestion charges might apply depending on the data you ingest." | |
| Reading note | Last updated September 23, 2026. The page does not mention a data lake or regions. | |
| Microsoft (2026h)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in The Sentinel Entitlement | Analytics tier | "Commitment tier pricing starts at 100 GB per day." |
| Free data | Security alerts from Defender products are free, while some raw Defender, Entra ID, and other logs are paid (summarized, not quoted). | |
| Reading note | Last updated May 6, 2026. No list prices. The page does not mention E7. | |
| Microsoft (2026i)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Vulnerability Shielding | Actions | Contains at-risk user accounts, hardens group policy settings, enforces Safeboot, or restricts access to sensitive data for at-risk devices (summarized, not quoted). |
| Licensing | "To use these actions, you need a Defender for Endpoint license." | |
| Reading note | Last updated June 23, 2026. The page does not mention vulnerability shielding. |
Rival vendors11 sources, 20 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Amazon Web Services (2025a)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in The Final Greyhound Standpoint, Rivals Compared, Microsoft Perception. Table 4 | Announcement | "The new pricing model introduces a free tier covering the first 10,000 findings per month" |
| Amazon Web Services (2025b)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Rivals Compared. Table 4 | Announcement | The agent asks the case submitter clarifying questions and gathers evidence across AWS data sources (summarized, not quoted). |
| Amazon Web Services (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Rivals Compared. Table 4 | Announcement | "Today, AWS announced the general availability of AWS Security Agent for on-demand penetration testing in six AWS Regions." |
| Amazon Web Services (n.d.)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Rivals Compared | Guide | "AWS Security Agent (now part of AWS Continuum) is a frontier agent that proactively secures your applications throughout the development lifecycle." |
| Reading note | The guide covers penetration testing, code review, threat modeling, and design review, and does not mention the SOC. | |
| Blaier, A. and Goh, J. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Rivals Compared | Post | "Currently in private preview. General availability in Q4FY26." |
| Reading note | Byline June 11, 2026, metadata modified August 19. Palo Alto Networks' fiscal year ends in July, so the stated quarter has passed. See Simkin (2026). | |
| Chaudhary, U. and Jagadeesh, N. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Pricing, Rivals Compared. Table 4 | Post | "For example, Google Security Operations will let you set token-budgets based on which agent you are using." |
| Reading note | The post's other figures are not used in this note. | |
| CrowdStrike (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Agents As Insiders, Staffing And Sourcing, Rivals Compared. Table 4 | Release | "Analysts pause or stop a Charlotte AI or AgentWorks running agent mid-task" |
| Release | Coordinated multi-agent investigations are generally available, and Agent Insider Threat Detection is listed for the fourth quarter (summarized, not quoted). | |
| Release | "Analysts approve, reject with a documented reason, or inspect the full agent trace." | |
| Release | AI credit visibility and control, under which admins see the credits every SOC agent consumes and can halt or cap it, is listed for the fourth quarter (summarized, not quoted). | |
| Fink, G. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Rivals Compared. Table 4 | Post | Names the Case Investigation, Cloud Posture, and Automation Engineer agents and the Cortex AgentiX platform (summarized, not quoted). |
| Reading note | The post gives no pricing or general availability dates for individual agents. | |
| Regalado, P. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Staffing And Sourcing, Rivals Compared. Table 4 | Post | "The Triage Agent helps evaluate and explain findings, so analysts can focus on what matters." |
| Reading note | The post gives no status for the agents and no pricing. | |
| SentinelOne (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Rivals Compared. Table 4 | Release | "Investigations utilize Singularity Credits during the trial, but customers are not charged and no payment method is required." |
| Release | "After the trial, customers can purchase Singularity Credits through partners, direct billing, and eCommerce." | |
| Simkin, S. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in Rivals Compared | Reading note | Palo Alto Networks' July 2026 Cortex release post. It gives no general availability label for frontier model support. |
Media3 sources, 4 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Lyons, J. (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in The Final Greyhound Standpoint, Rivals Compared, Microsoft Perception. Table 4 | Article | Google's Triage and Investigation agent is generally available, and the Threat Hunting and Detection Engineering agents are in preview (summarized, not quoted). |
| Article | Google claims the agent cut a typical 30-minute manual analysis to 60 seconds, a company claim not used in this note's analysis (summarized, not quoted). | |
| Office Watch (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in What Microsoft Announced | Article | Counts each separately named Ignite feature with its own stated status once (summarized, not quoted). |
| WindowsForum (2026)Re-read November 19, 2026, between 00:28 and 01:15 UTCUsed in No section. Retained for reference | Article | Describes Perception's cloud coverage, a claim not used because Microsoft's own pages name AWS, Okta, and Salesforce (summarized, not quoted). Updated November 18, 2026. |
Greyhound Research7 sources, 10 extracts
| Source, read, and used | Where in the source | What it says |
|---|---|---|
| Greyhound Research (2025a)Re-read November 19, 2026, on greyhoundresearch.comUsed in The Triage Agent | Recommendation 10 | "At Greyhound Research, we believe the next wave of cloud security maturity will be judged by autonomy, not headcount." |
| Recommendation 10 | Traditional measures such as alerts closed and coverage achieved do not reflect what AI-native security changes (summarized, not quoted). | |
| Recommendation 3 | "If you can't audit the algorithm, you can't own the risk." | |
| Greyhound Research (2025b)Re-read November 19, 2026, on greyhoundresearch.comUsed in Agents As Insiders | Greyhound CIO Pulse 2025 | "According to Greyhound CIO Pulse 2025, 69% of CIOs surveyed now say their top concern with AI agents is not performance or hallucination" but "silent privilege creep" (fragments). |
| Greyhound CIO Pulse 2025 | "According to Greyhound CIO Pulse 2025, 57% of CISOs in Global 2000 firms have cited agent identity ambiguity as a top-5 security concern for GenAI deployment." | |
| Greyhound Research (2025c)Re-read November 19, 2026, on greyhoundresearch.comUsed in The Sentinel Entitlement | Greyhound Standpoint | Microsoft's European Security Program is a strategic escalation in the platform wars, in which cybersecurity is no longer a revenue line but a loyalty lock (summarized, not quoted). |
| Greyhound Research (2026a)Re-read November 19, 2026, on greyhoundresearch.comUsed in Vulnerability Shielding | Synopsis | "When AI finds flaws faster, the scarce resource is the capacity to fix them." |
| Greyhound Research (2026b)Re-read November 19, 2026, on greyhoundresearch.comUsed in Pricing | Greyhound CIO Pulse 2026 | "Greyhound CIO Pulse 2026, across more than 1,000 enterprise technology leaders, finds 84 percent" "have shifted their focus from AI adoption to AI accountability" (fragments). |
| Greyhound Research (2026c)Re-read November 19, 2026, on greyhoundresearch.comUsed in The Strongest Objection | Reading note | The sentence quoted in The Strongest Objection sits in Section 1.1, The Deeper Read, of the November 15 note, which attributes it to Greyhound Research's public statement of November 13, 2026. |
| Greyhound Research (2026d)Draft re-read November 19, 2026. Not yet publishedUsed in The Final Greyhound Standpoint, Agents As Insiders, Microsoft Perception | Section 2.2, Agents As Insiders | Agents are insiders. Harm arrives by three routes, a sandbox stops only one, and narrow rights, approval, identity, and a quarterly review stop the other two (summarized, not quoted). |
Part 8Appendix, Sources, And Disclosures
Sources And References: Every Document Behind Every Claim
Every fact, figure, and quotation in this research note traces to one of the sources below, presented in Harvard reference format and grouped by class. Each source is also linked in the sentence that carries it. Microsoft’s and its rivals’ documents are each company’s account of its own products, prices, and terms, and prove what the company said rather than that the thing is true. The news reports what its authors were told. Standing pages that carry no publication date are marked n.d. All sources were re-read on November 19, 2026, between 00:28 and 01:15 UTC.
Microsoft announcements and blogs
Althoff, J. (2026) Introducing the First Frontier Suite built on Intelligence + Trust. The Official Microsoft Blog, March 9. Available at: https://blogs.microsoft.com/blog/2026/03/09/introducing-the-first-frontier-suite-built-on-intelligence-trust/
Gallot, H. (2026a) Rethinking security for the age of AI. The Official Microsoft Blog, July 27. Available at: https://blogs.microsoft.com/blog/2026/07/27/rethinking-security-for-the-age-of-ai/
Gallot, H. (2026b) Securing the agentic insider. The Official Microsoft Blog, October 19. Available at: https://blogs.microsoft.com/blog/2026/10/19/securing-the-agentic-insider/
Holeček, A. (2026) Security for the agentic era, powered by Microsoft Perception. Microsoft Security Blog, November 17. Available at: https://www.microsoft.com/en-us/security/blog/2026/11/17/security-for-the-agentic-era-powered-by-microsoft-perception/
Lefferts, R. (2026) Reimagining the SOC for the agentic era in Microsoft Defender. Microsoft Security Blog, September 23. Available at: https://www.microsoft.com/en-us/security/blog/2026/09/23/reimagining-the-soc-for-the-agentic-era-in-microsoft-defender/
Microsoft (2026a) Microsoft Ignite 2026 | Book of News. Microsoft News, November 17. Available at: https://news.microsoft.com/events/ignite/2026/book-of-news/
Microsoft (2026b) What’s new with Microsoft Perception at Ignite 2026. Microsoft Tech Community, Perception Blog, November 17. Available at: https://techcommunity.microsoft.com/blog/perception-blog/whats-new-with-microsoft-perception-at-ignite-2026/4571834
Microsoft (2026c) Ignite 2026: What’s new in Microsoft Sentinel. Microsoft Tech Community, Microsoft Sentinel Blog, November 17. Available at: https://techcommunity.microsoft.com/blog/microsoftsentinelblog/ignite-2026-whats-new-in-microsoft-sentinel/4571174
Nadella, S. (2026) Ignite 2026: Building the frontier ecosystem. The Official Microsoft Blog, November 17. Available at: https://blogs.microsoft.com/blog/2026/11/17/ignite-2026-building-the-frontier-ecosystem/
Microsoft product and pricing pages
Microsoft (n.d.a) Microsoft Perception. Microsoft Security. Available at: https://www.microsoft.com/en-us/security/business/microsoft-perception-agentic-system
Microsoft (n.d.b) Project Perception. Microsoft Security. Available at: https://www.microsoft.com/en-us/security/business/ai-powered-cybersecurity/project-perception-agentic-system
Microsoft (n.d.c) Microsoft Security Copilot pricing. Microsoft Security. Available at: https://www.microsoft.com/en-us/security/pricing/microsoft-security-copilot/
Microsoft documentation
Microsoft (2026d) Get started with Project Perception. Microsoft Learn, updated September 6. Available at: https://learn.microsoft.com/en-us/security/agentic-security/agentic-security-get-started
Microsoft (2026e) Triage Agent. Microsoft Learn, updated August 19. Available at: https://learn.microsoft.com/en-us/security/agentic-security/triage-agent
Microsoft (2026f) Microsoft Security Copilot Security Alert Triage Agent in Microsoft Defender (Preview). Microsoft Learn, updated July 2. Available at: https://learn.microsoft.com/en-us/defender-xdr/security-alert-triage-agent
Microsoft (2026g) Integrated Security Operations Center (ISOC) in Microsoft Defender. Microsoft Learn, updated September 23. Available at: https://learn.microsoft.com/en-us/defender-xdr/isoc-overview
Microsoft (2026h) Plan costs and understand pricing and billing for Microsoft Sentinel. Microsoft Learn, updated May 6. Available at: https://learn.microsoft.com/en-us/azure/sentinel/billing
Microsoft (2026i) Predictive shielding in Microsoft Defender. Microsoft Learn, updated June 23. Available at: https://learn.microsoft.com/en-us/defender-xdr/shield-predict-threats
Rival vendors
Amazon Web Services (2025a) AWS Security Incident Response now offers metered pricing with free tier. AWS What’s New, November 21. Available at: https://aws.amazon.com/about-aws/whats-new/2025/11/security-incident-response-metered-pricing-free-tier/
Amazon Web Services (2025b) AWS Security Incident Response now provides agentic AI-powered investigation. AWS What’s New, November 21. Available at: https://aws.amazon.com/about-aws/whats-new/2025/11/aws-security-incident-response-agentic-ai-powered-investigation
Amazon Web Services (2026) AWS Security Agent on-demand penetration testing is now generally available. AWS What’s New, March 31. Available at: https://aws.amazon.com/about-aws/whats-new/2026/03/aws-security-agent-ondemand-penetration/
Amazon Web Services (n.d.) What is AWS Security Agent (now part of AWS Continuum)?. AWS Security Agent User Guide. Available at: https://docs.aws.amazon.com/securityagent/latest/userguide/what-is.html
Blaier, A. and Goh, J. (2026) The Frontier AI SOC Has Arrived. Palo Alto Networks Blog, June 11. Available at: https://www.paloaltonetworks.com/blog/security-operations/cortex-frontier-ai/
Chaudhary, U. and Jagadeesh, N. (2026) FinOps for SecOps: How to optimize the agentic SOC for value. Google Cloud, July 10. Available at: https://cloud.google.com/transform/finops-for-secops-how-to-optimize-the-agentic-soc-for-value
CrowdStrike (2026) CrowdStrike Extends the Agentic SOC to Detect and Stop the Agent Insider Threat. Press release, November 4. Available at: https://ir.crowdstrike.com/news-releases/news-release-details/crowdstrike-extends-agentic-soc-detect-and-stop-agent-insider
Fink, G. (2026) The SOC Is Now Agentic: Introducing the Next Evolution of Cortex. Palo Alto Networks Blog, February 25. Available at: https://www.paloaltonetworks.com/blog/2026/02/soc-agentic-next-evolution-cortex/
Regalado, P. (2026) Defending at Machine Speed: Splunk Advances the Agentic SOC. Splunk Blog, June 2. Available at: https://www.splunk.com/en_us/blog/security/splunk-advances-the-agentic-soc.html
SentinelOne (2026) SentinelOne Opens Purple AI Agentic Investigation to All Customers, Bringing Frontier AI Directly Into the SOC. Press release, June 17. Available at: https://www.sentinelone.com/press/sentinelone-opens-purple-ai-agentic-investigation-to-all-customers-bringing-frontier-ai-directly-into-the-soc/
Simkin, S. (2026) What’s New in Cortex (July ‘26). Palo Alto Networks Blog, July 30. Available at: https://www.paloaltonetworks.com/blog/security-operations/whats-new-in-cortex-july-2026/
Media
Lyons, J. (2026) Google unleashes even more AI security agents to fight crims. The Register, April 22. Available at: https://www.theregister.com/security/2026/04/22/google-unleashes-even-more-ai-security-agents-to-fight-crims/5221298
Office Watch (2026) Microsoft Ignite 2026: only 76 of 215 new features are ready today. Office Watch, November 18. Available at: https://office-watch.com/2026/microsoft-ignite-2026-ready-today/
WindowsForum (2026) Microsoft Ignite 2026 Keynote: Copilot Agents, AI Security and Azure Launches Explained. WindowsForum, November 17. Available at: https://windowsforum.com/news/microsoft-ignite-2026-keynote-copilot-agents-ai-security-and-azure-launches-explained.452746/
Research this note builds on
Greyhound Research (2025a) Google and Wiz: Why AI-Native Security Will Redefine Multi-Cloud Trust. Research note, May 2. Available at: https://greyhoundresearch.com/google-and-wiz-why-ai-native-security-will-redefine-multi-cloud-trust/
Greyhound Research (2025b) Microsoft Build 2025: Making AI Agents More Capable and Secure. Research note, May 22. Available at: https://greyhoundresearch.com/microsoft-build-2025-making-ai-agents-more-capable-and-secure/
Greyhound Research (2025c) Microsoft’s Cybersecurity Initiative for EU Governments. Research note, June 7. Available at: https://greyhoundresearch.com/microsofts-cybersecurity-initiative-for-eu-governments/
Greyhound Research (2026a) Oracle AI Security – When AI Finds The Flaws Faster, Fixing Becomes The Control. Research note, October 29. Available at: https://greyhoundresearch.com/oracle-ai-security-when-ai-finds-the-flaws-faster-fixing-becomes-the-control/
Greyhound Research (2026b) The Front Door Race – Oracle, OpenAI And Five Rivals, And Why The Permission Becomes The Control. Research note, November 5. Available at: https://greyhoundresearch.com/oracle-openai-the-front-door-race-the-permission-becomes-the-control/
Greyhound Research (2026c) Microsoft 365 Outage – When The Suite Goes Dark, Restoration Order Becomes The Control. Research note, November 15. Available at: https://greyhoundresearch.com/microsoft-365-outage-when-the-suite-goes-dark-restoration-order-becomes-the-control/
Greyhound Research (2026d) Microsoft Azure Agent Infrastructure – When Agents Get Their Own Machines, Authority Becomes The Control. Research note, November 2026. Available at: https://greyhoundresearch.com/microsoft-azure-agent-infrastructure-when-agents-get-their-own-machines-authority-becomes-the-control/
Part 8Appendix, Sources, And Disclosures
Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright
The evidence in this research note is current through November 19, 2026, with an evidence cutoff of 01:15 UTC that day, when every source was last re-read, two days after Microsoft’s Ignite announcements. Any written response received from an organization named in it after that time is dated where it appears.
Independence Disclosure
Research under the Greyhound Research brand is independent, and no organization holds a right of approval over its content or publication. It is not commissioned, sponsored, or funded by any organization it covers. Greyhound Research may send a research note to companies named in it for factual review before publication. Any such review is confined to the correction of fact, and no viewpoint, conclusion, or recommendation is guided by those companies. All insights and recommendations are Greyhound Research’s alone, and editorial control remains entirely with Greyhound Research. Greyhound Consilium is the B2B thought-leadership arm of The House Of Greyhound and a sister brand to Greyhound Research. It builds on Greyhound Research’s independent research through collaborations that produce thought-leadership content, research, and practical toolkits for decision-makers. Every Greyhound Consilium publication names its partner on the face of the document. This research note is published under the Greyhound Research brand.
Financial Disclosure
Greyhound Research undertakes advisory and consulting engagements with enterprise buyers and technology providers, and no such engagement confers control over its research findings or its decision to publish them. For this research note, Greyhound Research has no financial interest in, and no commercial relationship with, Microsoft, Amazon Web Services, Google, Anthropic, or OpenAI to declare. Oracle is a client of Greyhound Research. Oracle is not a subject of this research note.
Intended Use
This research note is written for chief information security officers, SOC leads and security operations managers, procurement and vendor management leaders, chief information officers, heads of IT operations, finance leaders, and the press. It reads Microsoft’s security announcements at Microsoft Ignite 2026, the documentation and pricing pages beside them, and the published announcements of rival security vendors, to inform decisions on agentic triage, agent rights, vulnerability shielding, security pricing, SIEM entitlements, and SOC staffing and sourcing. This research note is not investment advice, a recommendation to buy, sell, or hold any security, or a view on any company’s share price or valuation. It is not legal advice, and its reading of contract and licensing matters is to be confirmed with counsel. Its assessments rest on the documents listed in Sources And References as they stood on November 19, 2026, and can change as the companies named publish more. Product availability, status, regions, features, and prices are described as each company published them through November 19, 2026, so confirm them with the vendor before acting on them.
Research Methodology
Greyhound Research’s analysis combines lived experience with public sources. The lived experience is years of continuously tracking the market, Greyhound Fieldnotes from ongoing advisory engagements with enterprise buyers and with technology providers, and Greyhound Pulse, the firm’s market trackers. Lived experience forms the judgment and never stands in for evidence. The public sources are the thirty-nine documents listed in Sources And References, in six classes, and every factual claim is verified against the document that supports it. The passages relied on are quoted in The Deeper Read of the section that uses them, and the other words taken from each source, the passages summarized rather than quoted and the reading notes, are set out in the appendix. A finding that something is not stated or not documented is bounded to the documents the text names, as they stood at 01:15 UTC on November 19, 2026. Statuses are given as each company states them.
Microsoft’s pages disagree on Perception’s status and on the terms of the SIEM offer for E5 and E7 customers, and this note gives each page’s version with its date rather than choosing between them. Microsoft’s figure of about 80% for the Triage Agent is a coverage claim from its own analysis and is not used as a measure of accuracy. Rival vendors are compared by product, on their own announcements, and no vendor’s figures are combined. The Greyhound CIO Pulse figures are cited as published in Greyhound Research’s research notes of May 22, 2025, and November 5, 2026, with their edition years. Earlier Greyhound Research positions are cited from its published notes or from its published statements, and where this note’s view has moved, the section says so and why. No composite figure is computed. No Greyhound Fieldnotes are quoted, and no factual claim in this research note rests on unpublished material.
The reading of the alert path, the agents’ powers, the shield’s life, the price groups, the readings in Exhibits 1 to 7 and Tables 1 to 7, the risks, the Greyhound CXO Runbook, the questions for Microsoft, the signals, and every term proposed for a contract are Greyhound Research’s analysis, not statements by Microsoft or any other company. Quotations were re-checked against the live source on November 19, 2026.
Authoring Disclosure
This material reflects the analysis, viewpoint and editorial judgment of Greyhound Research. The position presented is developed from Greyhound Research’s research, advisory work and accumulated institutional knowledge. Depending on the assignment, AI may be used to retrieve relevant material; structure, draft or polish prose from analyst-provided concepts, dictation, notes, rough drafts and source material; support secondary research and factual cross-checking; and assist with editing, condensation and consistency with Greyhound Research’s house style. AI does not independently establish Greyhound Research’s viewpoint, analytical position or conclusions. Material factual claims are verified against source evidence before release. Greyhound Research reviews and approves the final material and retains responsibility for it.
Copyright Policy
All content in this document and on the Greyhound Research website is protected by copyright law and may not be reproduced, distributed, transmitted, displayed, published, or broadcast without the prior written permission of Greyhound Research or, in the case of third-party materials, the prior written consent of the copyright owner of that content. You may not alter, delete, obscure, or conceal any trademark, copyright, or other notice appearing in any Greyhound Research content. Quotation of up to one hundred words is permitted where it carries attribution to Greyhound Research and a link to the published page. Reproduction, redistribution or republication beyond that extent, in whole or in part, by email or in any medium including websites, newsletters and intranets, requires prior written permission, which may be requested at the address below. The published version of each content piece on the Greyhound Research website carries sharing tools for readers who wish to share it.
For questions arising from these disclosures, including licensing requests, contact the Greyhound Research Community Relations Team at connect@thofgr.com. For media inquiries, write to media@thofgr.com.

Analyst In Focus: Sanchit Vir Gogia
Sanchit Vir Gogia, or SVG as he is popularly known, is a globally recognized technology analyst, innovation strategist, digital consultant and board advisor. SVG is the Chief Analyst at Greyhound Research (part of The House Of Greyhound), a Global, Award-Winning Technology Research, Advisory, Consulting & Education firm. Greyhound Research works closely with global organizations, their CxOs and the Board of Directors on Technology & Digital Transformation decisions.
Discover more from Greyhound Research
Subscribe to get the latest posts sent to your email.
