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 Oracle AI World, Oracle presented Fusion Claw function by function. Its four releases around the event named sixteen Claw-powered applications across finance, HR, supply chain and customer experience, of the twenty-five Oracle says are available. Seven of the sixteen were new names, and six of the seven are in HR, in pay, payroll, hiring and positions. Oracle’s HR release states that organizations retain responsibility for employment decisions. No release gives a date, region or price, and Oracle’s documentation places the applications in Release 26D.
Greyhound Research reads the announcements as the point at which Fusion Claw stops being one decision. Each application sits on a Fusion module with its own approvals, its own way back, its own calendar and, in HR, its own law, so each needs a named owner who signs for what it commits. Oracle’s releases describe every application with the point at which a person comes in. Its 26D documentation, where it exists, is more cautious than the releases in HR and customer experience, and more autonomous in finance. Five of the sixteen have a 26D page that describes them as Claw-powered, and none of the six new HR names has one yet.
The hardest questions sit in the signatures, the handoffs and the calendar. Where an approver runs an application under their own login and then approves its recommendation, the trail can show one identity on both sides. One approved action can cross three to five owners, and no outcome record yet links the steps. The update arrives in the busiest weeks of the year for payroll, finance and supply chain. The research note sets out a six-move Greyhound CXO Runbook, one move for each leader, with twelve exhibits, five tables and seven questions for Oracle.
In This Research NoteSection guide · AI drafted · Analyst reviewed
Sixteen sections, each with a short summary. Select a title to go straight to that section.
- 1What AI World Changed: Fusion Claw Stops Being One Decision
Oracle used AI World to present Fusion Claw function by function. Greyhound Research reads it as the point at which Fusion Claw stops being one decision, and sets the test: a named owner for every application.
- 2What Oracle Announced: Sixteen Applications, And The Weight Moved To HR
The AI World releases name sixteen applications, seven of them new names and six of those in HR. HR now has ten names, the most of any function, and finance still has two.
- 3How Far Each Application Is Written To Go: Oracle’s Words Against The Autonomy Dial
Oracle’s release wording writes four applications in full-auto terms, four to review and then execute, seven to stop at a recommendation, and one to produce a document. Where a 26D page exists, it can promise less than the release, or more.
- 4Finance: One Application, A Stake In Ten, And A Signature Test
Ledger is the one finance application, and ten of the sixteen descriptions name money. A Claw recommendation can reach its approver three ways, and in one of them the same identity prepares and approves.
- 5HR: Pay, Payroll, Hiring And Positions, From The Module Up
Each HR application lands on a Fusion module with a way back and a point after which the change has reached a person. The EU AI Act, New York City’s law, US state rules and the EU Pay Transparency Directive reach the decision applications first.
- 6Supply Chain: Commitments That Leave The Company, And What It Costs To Recall Them
Each of the three applications ends in a commitment outside the company, and each way back narrows once the other party has it. Greyhound Research sees supply chain as the function ready first.
- 7Customer Experience: Five Applications, And The Changes That Cascade
The Fusion modules beneath the five applications pass a change on by themselves, so the checkpoint belongs before the change. No customer has yet spoken on the record about running one.
- 8The Handoffs: One Decision, Several Owners
Four chains in which one application’s approved action becomes another function’s input. Each crosses at least three owners, and no outcome record yet links the steps.
- 9One Runtime, Four Owners: Budgets, Stops, And The People Around Them
Budgets per application, warning thresholds, auto-stop and a stop for all items map onto ownership by function. Table 4 sets out who signs, who co-signs and who holds each budget.
- 10The Calendar: Release 26D Meets Each Function’s Year
Applying Oracle’s cohort rule puts Release 26D in production on November 20, 2026, December 18, 2026, or January 15, 2027, and payroll environments take the first. It lands well for sales planning and in the busiest weeks for payroll, finance and supply chain.
- 11Risks To Watch: Where Claw Across Functions Can Go Wrong
Six risks, from one identity preparing and approving to cost and chains scattered across four owners. Each remedy sits with a role the buyer already has.
- 12The Greyhound CXO Runbook: Six Moves, One For Each Leader
Six moves, one for each leader, from an owner for every application and chain to one authority model, one total budget and one stop. Four of the six start now.
- 13Questions For Oracle: What Buyers Will Ask Next
Seven questions, each led by a buyer signal, on names and cohorts, run identity, pay criteria, outreach, ways back, linked records and production customers. Each maps to a Runbook move a buyer can make meanwhile.
- 14Fusion Claw By Function: The Final Greyhound Standpoint
Oracle has taken Fusion Claw into every function and described each application with the point at which a person comes in. Buyers who treat ownership as the control gain most, and ten signals would move the position.
- 15Sources And References: Every Document Behind Every Claim
Eighty-four sources in seven classes, from Oracle’s announcements, documentation, blogs and investor day to named media, the rules and official texts cited, and the research this note builds on.
- 16Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright
Independence first, including that no named organization was offered a pre-publication factual review. The Financial Disclosure states that Oracle is a client of Greyhound Research and that the engagement did not fund this note. Intended use, research method, authoring, and copyright terms follow.
What AI World Changed: Fusion Claw Stops Being One Decision
Oracle used AI World to present Fusion Claw function by function. Its release on Fusion Applications said twenty-five Claw-powered agentic applications “are now available across” its finance, HR, supply chain and customer experience suites, and the four releases it issued around the event named sixteen of them. “The next era of enterprise AI is all about delivering software that increasingly does the work, while people define the objectives and set the policies,” said Chris Leone, executive vice president, Applications Development, at Oracle.
The weight of what was new fell on people. Seven of the sixteen had not appeared under those names before, and six of the seven are in HR, in pay, payroll, hiring and positions. The applications “carry out human-approved actions,” said Lew Thompson, senior vice president, Applications Development, at Oracle, and the HR release states who answers for the result: “For employment-related workflows, organizations retain responsibility for employment decisions.”
Greyhound Research reads the AI World announcements as the point at which Fusion Claw stops being one decision. A CIO can switch agentic applications on, but cannot sign for a pay plan, an allocation of scarce materials, or a configuration change that every sales quote runs on. In its research report on Fusion Claw, Greyhound Research found that the first decision “belongs with the controller, the head of HR or the supply chain owner, with the CIO as countersignatory.” AI World has put an application in front of each of them, each on a Fusion module with its own approvals, its own way back, its own calendar and, in HR, its own law. Oracle’s Fusion Service Descriptions put the same weight on the customer: “You are responsible for implementing appropriate human oversight, approval controls, and safeguards.”
This is Greyhound Research’s third note from Oracle AI World. Its first found that review becomes the control when agents draft the setup, and its second that fixing does when AI finds the flaws faster. When agents reach every function, ownership becomes the control.
It is also the third piece on Fusion Claw from The House Of Greyhound. The first, Greyhound Research’s research report, found that autonomy arrives with a receipt. The second, The Delegation Dial, set out how much authority to hand Fusion Claw, process by process. Greyhound Consilium, the thought-leadership arm of The House Of Greyhound, developed that guide in partnership with Oracle. This note follows Fusion Claw into every function.
The test for buyers is whether every Claw application they switch on has a named owner who signs for what it commits, can show who prepared and who approved each action, and knows how that action comes back if it is wrong. Passing it depends less on the agents than on four things buyers control: owners, signatures, the HR start and the calendar.
What Oracle Announced: Sixteen Applications, And The Weight Moved To HR
Oracle’s releases from the event organize Fusion Claw by function. The HR release came first, at the HR Technology Conference in the week before AI World, and named seven applications. The supply chain release named three and the customer experience release five. The release covering all of Fusion gave eight examples across the four functions, among them Ledger, the only finance application in the set. This note calls the four the AI World releases.
The count matters because names are what buyers plan on. At Oracle’s investor day, Leone said “we are delivering 75 agentic applications,” adding that “25 of those are Claw enabled.” Across Oracle’s releases and its Help Center page for Fusion Claw, Greyhound Research counts twenty-four distinct names, with no single list of twenty-five. This note states the names as Oracle wrote them and infers no renames.
Before AI World, supply chain held the most names. The launch release named four applications, and the Help Center page lists fifteen capabilities, six of them in supply chain. Together they give seventeen distinct names, seven in supply chain, four in HR, four in customer experience and two in finance.
The AI World releases moved the weight to HR. Six of the seven new names are HR applications, and the seventh, Executive Briefing Document, is in customer experience. None of the seven appears in Oracle’s earlier releases reviewed for this note, of September 29, March 24, April 9 for HR, April 9 for customer experience and August 11, or on the Help Center page. HR now has ten names, the most of any function. Finance still has two, Ledger and Collector Workspace. Exhibit 1 sets the two counts side by side.
Table 1 sets out all sixteen, with what each works on in Oracle’s words, and Exhibit 2 shows where a person comes in. No release gives a date, region or price for any of them, and each function release says only that the applications are “available within” their suite. Oracle’s documentation places them in Release 26D: “Release 26D of Oracle Fusion Cloud Applications introduces 25 Claw-powered applications,” its Fusion Insider blog says. Table 2 shows how far that documentation reaches, application by application.
In fuller terms, in Oracle’s words, Ledger works on “close readiness” and “accounting exceptions” across “large volumes of accounting entries.” Workforce Orchestration works on workforce plans that “balance demand shifts, cost, and operational continuity.” Payroll Command Center works across “employee data, imports, timecards, retroactive changes, and historical patterns,” and Positions Command Center across “position, employee, recruiting, cost, and approval data.” Service Health Check works on “emerging service-performance issues” and “likely root causes,” Strategic Field Demand Planning sets “field capacity” against “customer demand,” and Executive Briefing Document draws on “account engagement, company news” and “revenue potential.”
Greyhound Research · Table 1 · Function ledger
Sixteen applications across four functions, and what each one works on
The applications in Oracle’s AI World releases by function, with what each works on in Oracle’s words.
Each function’s applications commit something different, so each needs its own owner.
Table 1 lists the sixteen Claw-powered applications in Oracle’s AI World releases by function, with what each works on, in Oracle’s words. Finance has one, Ledger. HR has seven: Workforce Orchestration, Positions Command Center, Candidate Sourcing, Recruiting Command Center, Workforce Compensation, Payroll Command Center, and Payroll Activity Workspace. Supply chain has three: Supply Planning Allocation, Shipment Load Planning, and Procurement Spend. Customer experience has five: Account Territory Growth Plan, Service Health Check, Strategic Field Demand Planning, CPQ Test-to-Production Setup & Configuration Migration, and Executive Briefing Document. Seven are marked as first named in the AI World releases, six of them in HR. Current through October 30, 2026.
Greyhound Research · Exhibit 1 · Where the weight fell
AI World moved the weight of Fusion Claw’s named applications from supply chain to HR, where six of the seven new names sit
Distinct Claw application names by function, before and after the AI World releases.
HR now has the most names, ten of twenty-four. Finance has two, and added none.
Exhibit 1 shows two donut charts of distinct Fusion Claw application names by function, before the AI World releases and after them, with a table of the counts. Before, across Oracle’s launch release and its Help Center page, there are seventeen names: seven in supply chain, four in HR, four in customer experience and two in finance. After, there are twenty-four: ten in HR, seven in supply chain, five in customer experience and two in finance. The releases added seven names, six in HR and one in customer experience. A note marks Oracle’s own count of twenty-five, and Leone’s statement that 25 of 75 agentic applications are Claw enabled. Greyhound Research analysis of Oracle sources.
The mix runs against the order buyers have been moving in. In the research report of October, Greyhound Research read the buyer record as split by function: “Finance and HR are cautious, operations is further along.” The new names now sit in the function where buyers have been most careful, and where a wrong action reaches a person rather than a pallet.
Greyhound Standpoint: At Greyhound Research, we believe Oracle has delivered breadth with care. An agentic application in every function is breadth, and Oracle’s releases describe every one of the sixteen with the point at which a person comes in. For fourteen, that point is an approval or authorization, for Payroll Command Center it is a review, and for Executive Briefing Document it is a briefing a person edits. For HR, it has written employer responsibility for employment decisions into its own release, which is the sentence an employment lawyer will look for first.
The consequence is that the first decision has changed shape. Choosing an application now means choosing an owner, because the person who approves a pay plan is not the person who approves a freight plan. CIOs should take Table 1 to their leadership team, ask each function head which applications they will answer for, and ask the Oracle account team which of the twenty-four names arrive in their cohort’s update.
How Far Each Application Is Written To Go: Oracle’s Words Against The Autonomy Dial
The research report of October set out the three autonomy levels Oracle named for Fusion Claw in its September briefing: work assistance, deep research, and full auto. Oracle’s technical brief for Claw names three points on its autonomy dial, “Human in the loop,” “Human in the lead” and “Autonomous execution.” Oracle’s release wording for each of the sixteen applications can be read against those levels, and Greyhound Research reads it as a statement of design intent. The wording falls into four groups.
Four are written in terms of authority given in advance. Ledger applies computation to “execution that is approved and authorized, within organizational policies and controls.” Shipment Load Planning takes “approved action towards execution plans that have been authorized.” Account Territory Growth Plan works “before taking authorized action.” Candidate Sourcing coordinates outreach “aligned with established recruiting processes and approvals.” None names a review step before the action. The authority exists before the run, which is how full auto is described.
Four name a review, then execute. Workforce Orchestration launches “actions that are approved through organizational review.” Service Health Check prepares actions “for review and then executes approved actions.” Strategic Field Demand Planning prepares resource actions “for review before executing approved changes.” CPQ Test-to-Production Setup & Configuration Migration assembles “a reviewable report” and executes “approved migration steps within provided authority.” These read as deep research with execution after approval.
Seven stop at a recommendation. Workforce Compensation prepares plans “for human review and approval,” Recruiting Command Center prepares next steps “for recruiter consideration and approval,” Positions Command Center routes corrective actions “for review and approval,” and the two payroll applications prepare follow-up “for review” or “for review and approval.” Supply Planning Allocation prepares plans “for teams to review, refine, and approve,” and Procurement Spend recommends corrective actions “for team review and approval.” The releases do not say that these applications carry the approved action out.
One produces a document. Executive Briefing Document assembles “an editable, meeting-ready briefing,” which is work assistance.
Two of the four applications written closest to full auto reach people outside the company: candidates, through Candidate Sourcing, and carriers, through Shipment Load Planning. Candidate Sourcing is the only HR application in that group, and its work, grouping candidates and coordinating outreach and campaigns to them, sits close to what the EU AI Act describes when it names systems used “to place targeted job advertisements, to analyse and filter job applications.” Section 4 returns to that.
The function leaders use the same language. In HR, the applications “carry out human-approved actions,” in Thompson’s words. In supply chain, they “carry approved actions into execution,” said S.Y. Shenoy, senior vice president, Applications Development, at Oracle. In customer experience, they “carry approved actions into execution,” said Kartik Raghavan, senior vice president, Applications Development, at Oracle.
The Help Center page reads differently. Its descriptions of the fifteen capabilities never say that a person approves or authorizes an action before it runs. The page describes Claw as a way “to create autonomous outcomes,” and the nearest it comes to an action is Collector Workspace, which can “execute policy-permitted collection actions.” The releases are where Oracle put the person, and a buyer should take that wording into the contract and the configuration rather than rely on the product page.
Laid out in Exhibit 2, each of the three larger groups mixes functions. In Oracle’s release wording, four of the five customer experience applications carry action into execution, against two of the seven in HR.
Greyhound Research · Exhibit 2 · How far each is written to go
Oracle’s own wording writes four applications in full-auto terms, and two of those four reach people outside the company
Each application placed by its AI World release wording, against the autonomy levels in the research report of October. Table 2 sets it against Oracle’s 26D documentation.
Candidate Sourcing is the one HR application written in terms of authority given in advance.
Exhibit 2 places the sixteen applications in Oracle’s AI World releases in four lanes, by how far Oracle’s release wording lets each go, matched to the autonomy levels in Greyhound Research’s research report of October, with each application’s function shown by color. Written in terms of authority given in advance, the full-auto fit: Ledger, Candidate Sourcing, Shipment Load Planning and Account Territory Growth Plan. Review, then execution, deep research with execution: Workforce Orchestration, Service Health Check, Strategic Field Demand Planning and CPQ Test-to-Production Setup & Configuration Migration. Recommendation only, deep research: Positions Command Center, Recruiting Command Center, Workforce Compensation, Payroll Command Center, Payroll Activity Workspace, Supply Planning Allocation and Procurement Spend. Document, work assistance: Executive Briefing Document. Each application carries Oracle’s own approval words. The exhibit reads release wording only. Table 2 sets it against Oracle’s 26D documentation, which describes Workforce Orchestration as read-only and Strategic Field Demand Planning as advisory. Greyhound Research analysis of Oracle’s descriptions.
Two words carry more than they show. “Approved” can mean that a person approved this action, or that a person approved a class of actions once. “Authorized” does not say who authorized the action, or when. The releases also name the approver in general terms, “teams,” “recruiter,” “organizational review,” “human review,” and each of those has to become a person in the buyer’s organization. Oracle’s runtime leaves the setting to the customer, and The Delegation Dial, Greyhound Consilium’s practical guide to Fusion Claw developed in partnership with Oracle, sets out how to choose it process by process.
Oracle’s documentation is the second record, and it does not always say what the releases say. Greyhound Research looked in Oracle’s Release 26D readiness guides for a feature page that describes each of the sixteen applications as Claw-powered and, as of October 30, 2026, found one for five: Ledger, Workforce Orchestration, Procurement Spend, Account Territory Growth Plan and Strategic Field Demand Planning. None of the six new HR names has one yet. Oracle’s Recruiting 26D guide does document a Candidate Sourcing Workspace, tagged as an AI agent feature, which “Transforms candidate sourcing into a proactive, agentic process,” and its Service Center guide a Service Operations Workspace agentic app, but neither page names Fusion Claw, and this note infers no renames. Collector Workspace, finance’s other Claw capability, has a Claw page, and it carries the only availability date Greyhound Research found in the Oracle documents reviewed: “Fusion Claw capabilities will be available for customer uptake starting 13th Nov CWB,” its 26D page says.
Where a page exists, it can promise less than the release, or more. Workforce Orchestration is released as launching “actions that are approved through organizational review,” and its 26D page says “The workspace is read-only” and “doesn’t perform HCM transactions such as transfers, terminations, requisition postings, or position updates.” Strategic Field Demand Planning is released as “executing approved changes,” and its page says “Recommendations are advisory.” Account Territory Growth Plan’s documented actions are to “create opportunities, add accounts to engineered sales campaign programs, and create follow-up tasks for at-risk accounts,” and Procurement Spend analyzes “invoice lines created in the last completed quarter.”
Finance runs the other way. Ledger’s page lets customers “allow policy-approved automatic corrections” and schedule outcomes “to run automatically,” and Collector Workspace’s says “Eligible actions can be executed automatically in accordance with organizational policies and controls.” The names moved to HR at AI World, and the documented autonomy sits in finance.
Every page also says what a customer must do first. Ledger needs the “Ledger Agentic Application Assistant” role, Procurement Spend two profile options, a new duty role and a permission group, and Workforce Orchestration a profile option and a privilege that Oracle grants the Human Resource Specialist role out of the box. New features that materially change the user experience “are shipped ‘disabled’ and require opt-in,” Oracle’s Fusion Insider team writes. Not every default is off, though. In Sourcing Command Center, another Claw capability, “outcomes are enabled by default,” Oracle’s 26D page says, and turning them off means disabling autonomous outcomes in the application’s settings. Table 2 sets the two records side by side.
Greyhound Research · Table 2 · Announced, documented, switched on
Five of the sixteen applications have a 26D Claw page, and the pages that document automatic execution are all in finance
What each AI World release says, what Oracle’s 26D readiness page says, and what switching the application on takes.
No documented application runs until switched on. That switch is the owner’s first signature.
| Application | In the AI World release | In Oracle’s 26D documentation | To switch it on |
|---|---|---|---|
| LedgerFinance | “execution that is approved and authorized” | Automatic, by policy“policy-approved automatic corrections”outcomes scheduled “to run automatically” | “Ledger Agentic Application Assistant” role |
| Collector WorkspaceFinance | Not in the AI World releases | Automatic, by policy“Eligible actions can be executed automatically”available “starting 13th Nov CWB” | Collections policy published in AI Agent Studio |
| Workforce OrchestrationHR | “actions that are approved through organizational review” | Read-only“The workspace is read-only” | A privilege and a profile option |
| No 26D Claw pageHR | Positions Command CenterCandidate SourcingRecruiting Command CenterWorkforce CompensationPayroll Command CenterPayroll Activity Workspace | ||
| Procurement SpendSupply chain | “for team review and approval” | AnalysisReads “invoice lines created in the last completed quarter” | Two profile options, a duty role and a permission group |
| No 26D Claw pageSupply chain | Supply Planning AllocationShipment Load Planning | ||
| Account Territory Growth PlanCustomer experience | “before taking authorized action” | Limited actions“create opportunities,” campaign enrollments, “create follow-up tasks” | Sales Manager or Sales VP role, and Claw outcomes configured |
| Strategic Field Demand PlanningCustomer experience | “for review before executing approved changes” | Advisory“Recommendations are advisory” | Duty roles and permission groups |
| No 26D Claw pageCustomer experience | Service Health CheckCPQ Test-to-Production Setup & Configuration MigrationExecutive Briefing Document | ||
Table 2 sets each application’s AI World release wording beside what Oracle’s Release 26D readiness pages say and what switching the application on takes. Five of the sixteen have a 26D page that describes them as Claw-powered: Ledger, documented with policy-approved automatic corrections and outcomes scheduled to run automatically; Workforce Orchestration, documented as a read-only workspace; Procurement Spend, documented as analysis of last quarter’s invoice lines; Account Territory Growth Plan, documented with actions to create opportunities, add accounts to sales campaign programs and create follow-up tasks; and Strategic Field Demand Planning, documented as advisory. Collector Workspace, finance’s other Claw capability, is documented to execute eligible actions automatically from November 13. Each documented application needs a role, a profile option, a policy or a configured outcome before it runs. No such page was found for the six new HR names, Supply Planning Allocation, Shipment Load Planning, Service Health Check, the CPQ application or Executive Briefing Document. Recruiting 26D documents a Candidate Sourcing Workspace, and Service Center 26D a Service Operations Workspace, without naming Fusion Claw. Readiness guides checked October 30, 2026.
Greyhound Research reads the two records together as proof that the setting is the buyer’s. The same application can be advisory in Oracle’s documentation and executing in its release, and none of the documented applications runs until someone assigns a role, sets a profile option or publishes a policy. That act is the owner’s first signature.
Greyhound Standpoint: Greyhound Research finds Oracle’s approval wording well matched to the stakes in most functions. The HR applications that touch pay and candidacy stop at a recommendation, and so do the supply chain applications that allocate scarce materials or redirect spend. The design puts the person where an employee or a supplier would feel a mistake.
The exception to watch is Candidate Sourcing. It is the one HR application written in terms of authority given in advance, and its work reaches people outside the company under a rule written for exactly that work. Function heads should replace each general approver in Exhibit 2 with a named role and deputy, decide for every application that executes whether “approved” means each action or a class of actions, and keep Candidate Sourcing on per-message approval until legal has mapped it. Because Oracle’s releases and its documentation describe the same applications differently, each function head should also check the 26D page and the default setting before switching an application on, and treat that switch as a decision they sign.
Finance: One Application, A Stake In Ten, And A Signature Test
Finance received the least at AI World and has the most to watch. Ledger is the only finance application in the AI World releases. On Oracle’s Help Center, it “analyzes account variances, evaluates account usage against controls, and examines account activity to explain anomalies and highlight issues requiring action before close,” and it “incorporates standard operating procedures (SOPs) to guide analysis and actions.” The page adds Collector Workspace, which can “execute policy-permitted collection actions,” so the one finance capability that acts beyond the books acts toward customers.
Oracle’s 26D documentation goes further than either. On Ledger’s 26D page, a user can “Review the planned execution path generated from the outcome prompt before work begins,” adjust it, “then confirm execution,” and schedule recurring outcomes “to run automatically at intervals such as hourly, daily, or on weekdays.” Customers “can allow policy-approved automatic corrections or retain exceptions for team review,” and for now “Automatic resolution is possible for exceptions relating to missing currency rates and inactive / end-dated segment values.” Ledger can also “govern authorized reversals or account-usage adjustments.” Collector Workspace’s Claw capabilities will be available for customer uptake “starting 13th Nov CWB,” and the workspace uses the existing Receivables approval rules and write-off controls: “It doesn’t override these controls,” its page says.
The general ledger beneath it sets the terms of any correction. “You cannot delete posted journals but you can copy or reverse them,” Oracle’s documentation says, and “You can only reverse a journal once.” A reversal is given a reversal period, which Oracle’s options can set as the same period, the next period or the next nonadjusting period, so a posting found wrong after the close can be reversed in a later period. Correcting the closed period itself means reopening it, and Oracle notes that “In general, periods aren’t reopened unless there are strong justifications and related approvals.” Because a journal can be reversed only once, a reversal that Ledger makes uses that journal’s one reversal.
Leone has described the work in plain terms. “Think of a ledger with 1 million or 2 million records that you need to go reconcile or you need to go analyze,” he told investors, naming the work of “a financial analyst in accounting” among the specialist roles Claw takes on. The foundation matters as much as the application. Grupo Bimbo’s Oracle work began around “the financial information,” and “After implementing Fusion, we have been able to bring on the different Supply Chain modules,” said Antonio Parra, global business technology leader at Grupo Bimbo. He welcomed the chance “to change the closing time from 16 days to three days,” and put the larger value in supply chain.
The stake in the other three functions. Ten of the sixteen descriptions name money in some form, and nine of those belong to HR, supply chain and customer experience: pay and payroll, positions and workforce cost, freight and spend, overtime and revenue. Mapped in Exhibit 3, every function’s applications reach the CFO’s numbers somewhere.
Greyhound Research · Exhibit 3 · The CFO’s stake
Ten of sixteen descriptions name money, nine of them outside finance
Each application whose own description names money, by the kind of money. Quoted words are Oracle’s.
The CFO is the second signer of the Claw estate, whichever function an application sits in.
| Application, and the money it names | The books | Pay and payroll | Spend and freight | Labor cost | Revenue |
|---|---|---|---|---|---|
| Ledger“accounting entries” | |||||
| Workforce Orchestration“cost” | |||||
| Positions Command Center“cost data” | |||||
| Workforce Compensation“budgets, pay guidelines” | |||||
| Payroll Command Center“potential payroll issues” | |||||
| Payroll Activity Workspace“pay, balances, and costs” | |||||
| Shipment Load Planning“transportation costs” | |||||
| Procurement Spend“savings opportunities” | |||||
| Strategic Field Demand Planning“overtime costs” | |||||
| Executive Briefing Document“revenue potential” | |||||
| Applications | 2 | 3 | 2 | 3 | 1 |
Exhibit 3 maps the applications in Oracle’s AI World releases whose descriptions name money, by the kind of money: the books, pay and payroll, spend and freight, labor cost, and revenue. Ten of the sixteen name money in some form, and nine of those sit outside finance: Workforce Orchestration, Positions Command Center, Workforce Compensation, Payroll Command Center and Payroll Activity Workspace in HR, Shipment Load Planning and Procurement Spend in supply chain, and Strategic Field Demand Planning and Executive Briefing Document in customer experience. Ledger is the one finance application. Greyhound Research marks where an approved action commits money and where the application informs a decision about it. A strip of all sixteen applications by function shows which name money, column totals count the applications for each kind of money, and the six whose descriptions name no money are listed beside the strip. Greyhound Research analysis of Oracle’s descriptions.
The signature test. The second finance question is about who signed. Segregation of duties asks that the person who prepares a transaction is not the person who approves it, and Fusion Claw changes who prepares. Oracle’s technical brief for Claw says what each run records: “The outcome receipt records the requester, owner, executor, and approver, plus the governing policy,” with the data sources, the linked transactions, a verification ID and a digital signature. The identity rules beneath it were set out in July, before Claw launched, by Oracle’s Fusion Development Center of Excellence, writing about AI Agent Studio, where Claw’s autonomous outcomes are configured. “Every action carries the running user’s Fusion token,” it wrote, and a builder can schedule a Workflow Agent Team “to run under a service account, but scheduling requires an additional privilege.” Its rule of thumb is to “keep builder privileges, run access, and service-account access separate.” Greyhound Research reads those rules as the starting point for Claw runs, which Oracle has not yet documented separately.
That gives three patterns. If the approver runs the application under their own login and then approves its recommendation, the work was prepared under the approver’s token and approved by the approver, and the trail can show one identity on both sides.
If one person runs it and a different, named approver approves, the trail shows two identities and the separation holds. If it runs on a schedule under a service account, the service account prepared the work, the trail shows the service account and the approver, and the account’s privileges have to be kept apart from the approver’s. The receipt makes each pattern visible, because it names the requester and the approver. What Oracle has not documented is the identity a Claw run executes under, and how a receipt maps to Fusion’s own audit records, and Greyhound Research found neither in the Oracle documents reviewed.
A fourth pattern sits in finance. Collector Workspace reads incoming customer email from a mailbox that “is accessed using the user context of the administrator who configures the email account,” its 26D page says, so, in Greyhound Research’s reading, the identity on that access belongs to neither the collector nor an approver.
Greyhound Research raised the identity question before Claw existed. Its research report on Oracle AI Agent Studio asked whether each agent operates “as a distinct principal under least privilege,” or borrows “the entitlements of a human user.” For agents a person runs in AI Agent Studio, Oracle’s answer is on the record: the action carries the running user’s token. For Claw, the receipt now names the requester and the approver, and the identity a run executes under is still to be documented.
Oracle already sells the control that checks this for people. Its Advanced Access Controls use “Access-model filters” that “designate roles or privileges that, individually or in combination, would allow an individual user to complete risky behaviors,” in Oracle’s words. An Oracle Cloud Infrastructure blog says “agent action sets may be mapped to the same SoD policy model used for human roles” where Oracle Risk Management, Oracle Access Governance or equivalent controls are implemented and configured, and calls its examples reference patterns rather than built-in features. Native analysis of agent roles was not found in the Oracle documentation reviewed. Exhibit 4 sets out the first three patterns and what each shows an auditor.
Greyhound Research · Exhibit 4 · The signature test
A Claw recommendation can reach its approver three ways, and in one of them the same identity prepares and approves
Who prepares and who approves under each way of running a Claw application, and the control Greyhound Research proposes.
Bar self-approval where money moves. Hold agent roles to the rules people follow.
| Way it runs | Who prepares | Who approves | Segregation | Control proposed |
|---|---|---|---|---|
| The approver runs it, then approves | AThe approver | AThe approver | One identity | Bar self-approval in payroll, procurement, and the close |
| One person runs it, another approves | AThe running user | BA different approver | Separated | Keep run and approve rights in different roles |
| It runs on a schedule | AA service account | BA named approver | Separated if privileges stay apart | Hold service-account and agent roles to the same rules |
Exhibit 4 sets out three ways a Claw recommendation can reach an approver, beneath two lines from Oracle: from its Fusion Development Center of Excellence, writing about AI Agent Studio in July, that every action carries the running user’s Fusion token, and from its Fusion Claw technical brief, that the outcome receipt records the requester, owner, executor and approver. In the first pattern, the approver runs the application and then approves its recommendation, so one identity prepares and approves. In the second, one user runs it and a different approver approves, so the two are separated. In the third, it runs on a schedule under a service account, and a named approver approves, so the two stay separate only if their privileges do. For each, the exhibit names the control Greyhound Research proposes, from barring self-approval in payroll, procurement and the close to holding service-account and agent roles to the same rules as people. A note records what Oracle has not documented for Claw: the identity a run executes under, how a receipt maps to Fusion’s audit records, and native segregation of duties analysis of agent roles. Based on Oracle’s blogs, technical brief and Advanced Access Controls documentation. The readings are Greyhound Research’s.
Greyhound Standpoint: Greyhound Research holds that the CFO is the second signer of the Claw estate, whichever function an application sits in, and that the first thing the CFO should sign is the signature rule itself. Nine applications outside finance name money, and if Claw runs follow the identity rule Oracle has published for AI Agent Studio, any of them run by its own approver would prepare work under the approver’s token unless someone decides otherwise. For payroll, procurement and the close, that is the pattern an external auditor tests first.
Controllers should bar self-approval of Claw recommendations in the processes that move money, run scheduled outcomes only under service accounts whose privileges are kept apart from approvers’, and bring agent and service-account roles into the same segregation of duties rules as people. They should also plan the reversal window for Ledger, because a posting found wrong after the close is reversed in a later period, correcting the closed period needs a reopening Oracle reserves for strong justifications, and a reversal Ledger makes uses the journal’s only one. Before Collector Workspace’s Claw capabilities become available, which Oracle dates “starting 13th Nov CWB,” the controller should name the administrator under whose user context it reads the collections mailbox, and approve the collections policy it executes.
HR: Pay, Payroll, Hiring And Positions, From The Module Up
The six new HR applications reach the parts of employment where a decision changes someone’s livelihood. Workforce Compensation “develops recommendations for fair, budget-aligned compensation plans by evaluating options against employer-defined criteria, budgets, pay guidelines, and relevant equity considerations.” Payroll Command Center “works to identify potential payroll issues and improve run readiness.” Recruiting Command Center prepares next steps across “job, candidate, interview, feedback, offer, and demand data.” Candidate Sourcing coordinates “outreach, campaigns, and follow-up communications.” Positions Command Center “works to avoid potential disruptions from position changes.” Payroll Activity Workspace sets “current pay, balances, and costs” against prior periods.
Oracle has written the governing sentence into the release itself. Organizations “retain responsibility for employment decisions and can establish appropriate human oversight, approval processes, and controls consistent with their policies and applicable legal requirements.” Greyhound Research reads that as the right default. The application recommends. The employer decides, and the employer answers. Leone put the payroll outcome in the plainest words, to “pay people accurately on time within policy,” as he told investors.
What sits beneath each application. Each of the seven lands on a Fusion HCM process with its own approvals and its own way back, and Oracle’s documentation describes them precisely. None of the six new names has a 26D page that describes it as Claw-powered yet, so what follows is the module each will meet, read from Oracle’s documentation of the module itself.
Workforce Compensation lands on the compensation plan. Managers and specialists use worksheets “to propose compensation, performance ratings, and promotions,” in Oracle’s description, and a submitted worksheet goes to the “Manager one level above the submitter in the plan hierarchy,” until the plan is “Fully Approved.” The same notification table bypasses the submission notification where the “Submitting manager is the highest-level approver in the hierarchy,” which is the pay chain’s own version of the signature test. Approved changes reach salary records and payroll through a transfer process, and only changes in “Approved plans” are transferred. The transfer can be undone in the system. “To reverse the transfer of data from the workforce compensation plan to HR, you run the Back Out Workforce Compensation Data process,” Oracle writes. What Greyhound Research reads as beyond any back-out is a pay change an employee has already been told about.
Payroll Command Center lands on the payroll cycle, which runs from readiness through calculation and prepayments to payment, and the cycle has a hard line in it. Before a payment is issued, an incorrect run can be rolled back. After it, Oracle’s US payroll documentation is direct: “Once the process issues a payment, you can’t roll it back,” and reversals are for correcting “payroll calculations after a payment has been issued to the employee.”
The two leave different records. A reversal “generates a negative run result to offset the original run result, and maintains an audit trail,” while the Roll Back action negates the results “and leave no audit trail.” Payroll Activity Workspace, which looks at variances after a run, lands on payroll results, balances and costing, and costing has its own line. Once the transfer to the general ledger runs in final mode, Oracle says, “you can’t roll it back.”
That makes the timing of Payroll Command Center’s checks the whole of their value. A finding acted on before prepayments costs a rerun. The same finding after payment costs a reversal, a correction in the next run and, for the employee, a payslip that was wrong. And a correction made by rolling back leaves no audit trail in payroll, so the Claw outcome’s own record may be the only account of what the application found and what a person did about it.
Positions Command Center lands on position management, where one approval reaches many people. “All active assignments that are synchronized from position automatically inherit changes from the position,” Oracle’s documentation says. An approved corrective action on a position changes every assignment synchronized to it. The way back is to change the position again, which the synchronized assignments inherit in turn, and the change has reached people once every synchronized assignment has inherited it.
Recruiting Command Center and Candidate Sourcing land on Oracle Recruiting, where the offer has its own gates. An offer moves through states such as “Pending Approval,” “Approved,” “Extended” and “Accepted,” in Oracle’s documentation, and an offer “can be submitted for approval, then extended to the candidate.” Oracle’s existing AI outreach feature, in Recruiting update 26C, draws the line at the first message. Its AI-generated emails to candidates won’t be sent “until you select Send,” a follow-up can be scheduled “to be automatically sent out,” and when a candidate replies, “the AI agent reads the email and responds if it’s able.” The line holds for the first message, not for the conversation that follows. In update 26D, Recruiting’s Candidate Sourcing Workspace, an AI agent feature that does not name Claw, recommends “follow-up, shortlisting, profile-refresh campaigns, employee-referral campaigns, and event promotion,” and its outreach page adds “Automated follow-ups.” The Candidate Sourcing release does not say whether a person presses Send before its campaigns go out.
Workforce Orchestration is the one HR application with a 26D page, and that page draws a narrower line than the release. The release describes “actions that are approved through organizational review.” The page says the workspace “summarizes, analyzes, and recommends,” that suggested actions “aren’t automatically executed,” and that “Communications are suggestions for preview or copy; they aren’t sent from the workspace.” As documented for 26D, its plans reach people only through the HCM transactions a person then runs, such as transfers, terminations, requisition postings or position updates, and the point of no return sits in those.
Set out in Exhibit 5, each HR application has a path back in the system, and each also has a point after which the change has reached a person.
Greyhound Research · Exhibit 5 · HR, from the module up
Every HR application has a way back in Fusion, and every one has a point after which the change has reached a person
The seven HR applications in Oracle’s AI World releases, the way back in Oracle’s documentation, and the point of no return.
Act on payroll findings before prepayments. After payment, Oracle documents only a reversal.
Exhibit 5 draws a track for each of the seven HR applications in Oracle’s AI World releases, from the way back in Oracle’s documentation to the point after which the change has reached a person. Payroll Command Center, a check: roll back before payment and reverse after it, with no way back for the person once a payment is issued. Payroll Activity Workspace, a check: costing can be undone until it is final, after which Oracle says you can’t roll it back, and the point of no return is when costs post to the ledger. Positions Command Center, a check: change the position again, until synchronized assignments inherit the change. Workforce Compensation, a decision: the Back Out process reverses the transfer, until the employee is told the new pay. Recruiting Command Center, a decision: an offer is approved before it is extended, until it reaches the candidate. Candidate Sourcing, a contact: Oracle’s existing AI emails wait until a person selects Send, though its AI agent can answer replies, until a message is sent. Workforce Orchestration, a decision: its 26D workspace is read-only, and the point of no return is when a person runs the transfers or position updates it suggests. None of the six new names has a 26D Claw page yet, so each track describes the Fusion module beneath the application. The point-of-no-return readings are Greyhound Research’s.
Checks first, decisions later. The seven divide into checks and decisions, and the division sets the order of adoption. Payroll Command Center, Payroll Activity Workspace and Positions Command Center check: their findings go to a payroll or HR operations lead, and a wrong finding costs a second look if it is caught before the line in Exhibit 5. Workforce Compensation, Recruiting Command Center, Candidate Sourcing and Workforce Orchestration work on decisions and contacts that reach a person, and those are also the applications the law reaches first. Oracle expects the same order. Applications “that require customers to put a lot of standard operational procedures, policies and control around the AI” will, Rondy Ng told TechTarget, “come after they have built up enough confidence in the solution.”
The rules that reach each application. HR is the one function where specific laws already name the work these applications do. The EU AI Act lists as high-risk, in Annex III point 4(a), systems used “for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates,” in the Act’s text. Point 4(b) covers systems used “to make decisions affecting terms of work-related relationships, the promotion or termination of work-related contractual relationships, to allocate tasks based on individual behaviour or personal traits or characteristics.”
Those obligations now apply from December 2, 2027, under the amended Article 113, as amended by Regulation (EU) 2026/1744. Before using such a system at the workplace, employers “shall inform workers’ representatives and the affected workers,” under Article 26(7). The same article answers the owner question in law. Deployers “shall assign human oversight to natural persons who have the necessary competence, training and authority,” keep the system’s logs “for a period appropriate to the intended purpose of the high-risk AI system, of at least six months,” and, where a system is used to “make decisions or assist in making decisions related to natural persons,” must “inform the natural persons that they are subject to the use of the high-risk AI system.” In Greyhound Research’s reading, where an HR decision application is a high-risk system under the Act, the named approver this note recommends becomes a legal duty from December 2027, and the outcome receipt becomes a record the employer has to keep.
New York City’s law is already in force. It defines an automated employment decision tool as a process that “issues simplified output, including a score, classification, or recommendation, that is used to substantially assist or replace discretionary decision making,” in the city’s code, and an employment decision as one “to screen candidates for employment or employees for promotion within the city.” In Greyhound Research’s reading, outreach that screens no one may fall outside it, and grouping that ranks candidates may not. Under the city’s rules, that includes using an output “to overrule conclusions derived from other factors including human decision-making.”
A covered tool needs “a bias audit within one year of the use of the tool,” and candidates get notice “10 business days prior to use,” according to the city’s enforcement agency. The state comptroller found that the enforcing department’s “complaint handling and enforcement are inadequate,” and the department “agreed to adopt the majority of the recommendations in part or in full,” which points to tighter enforcement, not looser.
US state rules are arriving in sequence. Illinois makes it a civil rights violation “to use artificial intelligence that has the effect of subjecting employees to discrimination,” and requires notice to employees, from the start of 2026. Colorado’s replacement law, its attorney general says, goes “into effect January 1, 2027,” and HR Dive reports that it reaches consequential decisions “such as hiring and compensation determinations,” with notice, a description of the tool’s role in an adverse decision “within 30 days of the decision,” and the option to request “meaningful human review and reconsideration.” California’s privacy agency requires businesses to meet its automated decisionmaking requirements “by January 1, 2027,” for decisions that include employment.
Mapped by application in Table 3, the rules fall most heavily on the decision applications, and least on the payroll checks.
Greyhound Research · Table 3 · The rules that reach each HR application
The rules reach HR decision applications first, payroll checks least
Each HR application, the rules that may apply, and what to settle before the first run. Greyhound Research’s reading, to be confirmed with counsel, not legal advice.
Map every decision application to the law wherever the employer hires or pays, before it runs.
| Application | Rules that may apply | Settle before the first run |
|---|---|---|
| Candidate SourcingOutreach and campaigns | EU AI Act Annex III 4(a) · Dec 2, 2027New York City Local Law 144, where its output is used to screen candidatesIllinois · in forceColorado, California · Jan 1, 2027 | Whether a person approves each message before it is sentBias audit and notice where New York City law applies |
| Recruiting Command CenterCandidates “ready to advance” | EU AI Act Annex III 4(a)New York City Local Law 144, where its output carries the decisionUS state rules, as above | Whether recruiters can see why a candidate is readyNotice and audit duties by location |
| Workforce CompensationPay plans against criteria and budgets | EU Pay Transparency Directive · Jun 7, 2026EU AI Act Annex III 4(b)Colorado, California · Jan 1, 2027 | Objective, gender-neutral criteria behind each recommendationWorker representatives informed |
| Workforce OrchestrationWorkforce plans, read-only in 26D | EU AI Act Annex III 4(b), where its recommendations allocate tasks or decide termsWorks council rights, where they exist | Consult worker representatives before the first plan |
| Positions Command CenterCorrections that assignments inherit | EU AI Act Annex III 4(b), where changes decide terms or termination | Which changes need a second approver, by assignments affected |
| Payroll Command Center, Payroll Activity WorkspaceChecks on a payroll run | Data protection and payroll law, not the AI rules above | Which findings a person acts on before prepayments |
Table 3 sets out the rules that may reach each HR application in Oracle’s AI World releases, with their dates, in Greyhound Research’s reading, to be confirmed with counsel and not legal advice. Candidate Sourcing and Recruiting Command Center sit close to the EU AI Act’s Annex III 4(a) on recruitment, from December 2, 2027, New York City Local Law 144 where their output is used to screen candidates or carries the hiring decision, Illinois, in force, and Colorado and California from January 1, 2027. Workforce Compensation meets the EU Pay Transparency Directive, Annex III 4(b), and Colorado and California. Workforce Orchestration, read-only in 26D, and Positions Command Center meet Annex III 4(b) where their recommendations or position changes decide terms, and works council rights where they exist. The payroll applications meet data protection and payroll law rather than the AI rules. For each, the table states what to settle before the first run. Current through October 30, 2026.
Pay transparency: the risk and the opening. Workforce Compensation’s “relevant equity considerations” meets a law that asks for exactly that. The EU Pay Transparency Directive, which member states had to transpose by June 7, 2026, requires pay structures that allow an assessment of comparable work “on the basis of objective, gender-neutral criteria agreed with workers’ representatives where such representatives exist.” Employers with 250 or more workers report on the pay gap by June 7, 2027 and every year after, and those with 150 to 249 by the same date and every three years after. A gap of “at least 5 %” in a category that the employer “has not justified” on objective, gender-neutral criteria, and has not remedied within six months of reporting, triggers a joint pay assessment.
Oracle’s compensation page says its software can “Support pay equity and fair compensation,” and its datasheet lists “Pay discrimination analysis.” No Oracle document reviewed mentions the directive. The opening is in the receipt, and the directive asks for the same thing in its own words: employers “shall make easily accessible to their workers the criteria that are used to determine workers’ pay.” If the record of each Workforce Compensation outcome shows the criteria the recommendation weighed and the reason a manager departed from it, that record becomes the employer’s best evidence that pay decisions rest on objective criteria. If it does not, the application produces recommendations the employer will have to explain without them.
What HR customers said. HR customers at the event put the conditions first. Matt McDonald, director of HR technology at IHG Hotels & Resorts, described on a customer panel at Oracle AI World how IHG’s legal team weighs use cases, “less about what it does and more about intent and how you will use it.” An agent “just for interview scheduling” drew “Great, do it.” Talent Best Match drew the question “Where’s the human in the loop?” He said IHG’s leadership had identified talent acquisition as “a key area for step change.”
Doug Johnivan, vice president of operations and apps support at Acosta Group, gave the lesson on legal: “get them involved from day one, because they will slow you down, unfortunately.” He described “a lot of churn and time that we spent with legal around job postings” that “had to have language around AI in them.” Daniel Passarge, head of HR IT systems at Deutsche Bahn, said “Start with data cleansing and look into your data.” Kelli Scroggin, director of HRIS and analytics at UL Solutions, put it shortest: “nothing has no lift.” None of the four described running a Claw HR application, and their lessons apply to each of them.
Spectrum, which ChannelLife reported runs HR for roughly 100,000 people on Oracle HCM, drew the same line for its own AskHR agents. “If the response can be repeatable, governed, we let AI take care of it,” said Derek Dreyfus, vice president of HR technology at Spectrum. “If it’s requiring a pay change or an exception to a process… we keep that.”
Greyhound Standpoint: Greyhound Research considers HR the function where Claw should move last toward autonomy and can start first on checks. Oracle’s releases write five of the seven HR applications to stop at a recommendation, the one HR page in its 26D documentation describes a read-only workspace, and the release writes employer responsibility into Oracle’s own words. The modules beneath them have clear approval gates and documented ways back. The checks before a payroll run are high-volume, rule-bound work where a finding acted on in time costs a rerun rather than a wrong payslip.
The decisions carry the law with them. CHROs should start with Payroll Command Center, Payroll Activity Workspace and Positions Command Center when they reach the estate, since none has a 26D Claw page yet, act on payroll findings before prepayments, and require a second approver for position changes that reach many assignments. Before Workforce Compensation, Recruiting Command Center or Candidate Sourcing runs, general counsel should map each to the rules in Table 3 for every location the employer hires or pays in. The compensation owner should insist that each recommendation records the criteria it weighed, because that record is either the employer’s evidence of fair pay or a gap in it.
Supply Chain: Commitments That Leave The Company, And What It Costs To Recall Them
Supply chain held the most names before AI World and is where, in Greyhound Research’s reading, buyers have moved furthest with agents. It added no new name in the AI World releases. Its three applications each work toward a commitment to someone outside the company. Supply Planning Allocation “determines how to allocate scarce materials and capacity across competing demand.” Shipment Load Planning “works to reduce transportation costs and meet service level requirements.” Procurement Spend “works to improve spend control,” “detecting maverick spend and contract leakage.”
“As supply chain conditions shift, leaders need a connected view across product, planning, procurement, logistics, and cost to make faster, better-informed decisions,” said Shenoy. The applications “reason across Oracle Fusion Applications data, evaluate options and trade-offs, coordinate the work required to meet business objectives, and carry approved actions into execution,” he added.
What sits beneath each application. Supply Planning Allocation lands on Fusion supply planning, which already offers a choice the Claw release does not take. “You can either manually review and release each supply chain planning recommendation or automatically approve and release them,” Oracle’s planning documentation says. The Claw application is described at the stricter setting: plans “for teams to review, refine, and approve.” Once an order is released, cancelling it “entails releasing an order quantity of zero,” in the same documentation. Elsewhere in Fusion, allocation rules in backlog management can let the plan “steal supply” for high-priority nodes “from lower-priority nodes,” Oracle notes. That is the kind of decision a sales leader will want to see before a key customer comes up short.
Shipment Load Planning plans the loads that carriers are then asked to carry. Where those loads are executed through Oracle Transportation Management, a tender is a commitment with a clock. “When a shipment is tendered, notice is sent to a service provider with an offer to carry the shipment,” Oracle’s documentation says, and a timer starts. A tender can be withdrawn “before or after the service provider accepts the tender,” in Oracle’s words, and withdrawal “Re-tenders the shipment to the next higher cost service provider” where the shipment can be tendered automatically. The way back exists, and it can cost more freight than the plan saved. For some shipments, Oracle’s tender page says, Oracle Transportation Management “automatically tenders the shipment for you,” so the commitment to a carrier can already be made without a person. Oracle’s own Claw example points the other way. Of “154 proposed loads, 149 are sent downstream to the supply plan and five are held for further review,” its technical brief says, under the line “Fully autonomous does not mean unsupervised commitment.”
Procurement Spend, as Oracle documents it for 26D, stops before a supplier is reached. Its page analyzes “invoice lines created in the last completed quarter,” and users “consider recommended next steps before taking follow-up action.” Where a follow-up changes a purchase order, it reaches a supplier. On its enhanced Redwood change order page, Fusion lets a requester withdraw change orders “with statuses such as Pending Supplier Acknowledgement, Pending Approval, Pending Signature Preparation, or Pending Buyer Action.” Every status Oracle names is a pending one, and Greyhound Research found no Oracle description of withdrawing a change a supplier has already acknowledged, which then becomes a conversation with the supplier rather than a withdrawal in the system.
The customer voices on Claw itself. In the sources Greyhound Research reviewed, the customers at the event speaking about Claw itself came from supply chain, and from one company. Rob Dombkowski, senior director of business technology at Grupo Bimbo, said, in TechTarget’s report, that he attended a session on the new Fusion Claw SCM product, where a demo of an app that optimizes trucking capacity impressed him, and that “It really started giving me something tangible quickly that the business can get a benefit from.” He described the work it could replace: “Right now, the guy has to go and mine that data across 30 days to figure out why we were short on fulfillments.”
Parra, in diginomica’s account of his reaction to Oracle’s Claw announcements, spoke in the future tense. “It is going to be simplifying our work significantly,” he said, and “We are going to be getting rid of many of the manual applications that we have to put together and maintain.” On the supply chain side, he said, the cost points a company can take out are “hundreds of millions of dollars on an annual basis.”
Those are a buyer’s reaction and plans, not a production result. They still show why supply chain is ready. Its work is measurable, its data already sits in planning and logistics systems, and its gains are counted in freight, fill rates and spend. Leone illustrated the scale of action with Oracle’s own data-center logistics: “We can do and transact that 20 million movement of inventory around the world to get inventory to the right data centers,” he told investors.
Exhibit 6 follows each application from where the person comes in to the commitment, and on to the way back.
Greyhound Research · Exhibit 6 · Supply chain commitments
Each supply chain application ends in a commitment outside the company, and each way back narrows once the other party has it
Where the person comes in, the commitment outside the company, and the way back in Oracle’s documentation.
Hand over the analysis. A named planner or buyer releases each external commitment.
Exhibit 6 follows the three supply chain applications in Oracle’s AI World releases from where Oracle’s description puts the person, to the commitment outside the company, to the way back in Oracle’s documentation. Supply Planning Allocation stops at plans teams review, refine and approve, though Fusion planning can also release automatically, and its commitment reaches customers and plants that receive less of a scarce material. Shipment Load Planning takes approved action on authorized plans, and where its loads run through Oracle Transportation Management the commitment is a tender to a carrier, which can be withdrawn before or after acceptance but, where tendering is automatic, re-tenders at higher cost. Procurement Spend, documented for 26D as analysis of last quarter’s invoice lines, reaches suppliers only where a follow-up changes a purchase order, and change orders can be withdrawn in pending statuses. Each row names the signer Greyhound Research proposes. Greyhound Research analysis of Oracle’s releases and documentation.
Greyhound Standpoint: Greyhound Research sees supply chain as the function ready first, and AI World added the evidence it lacked: a customer on the record reacting to a Claw supply chain session, with a problem measured in days and dollars. The applications match the work, two of the three stop at a recommendation a person approves, Oracle has chosen the stricter release setting that its own planning software makes optional, and the one supply chain application with a 26D page, Procurement Spend, analyzes invoices and recommends.
The care belongs at the moment a commitment reaches the other party. A tender withdrawn after acceptance can re-tender at higher cost, and the change-order statuses Oracle names for withdrawal are all pending ones. Chief supply chain officers should hand over the analysis first, keep the release of tenders, allocations that leave a customer short, and supplier change orders with a named planner or buyer, and bring sales into any allocation that steals supply from one customer for another. Before Shipment Load Planning feeds Oracle Transportation Management, they should check which shipments it already tenders automatically. The Delegation Dial sets out the starting rule for a first process.
Customer Experience: Five Applications, And The Changes That Cascade
Customer experience has the second-longest list in the AI World releases and, in their wording, the furthest reach into execution. Its five applications work on how a company sells to and serves its customers.
On the Help Center, Service Health Check helps service managers “uncover gaps in knowledge, recommendations, guardrails, workflows and intent routing.” Account Territory Growth Plan gives sales leaders “a prioritized growth plan for every account.” Strategic Field Demand Planning “helps execute resource actions to improve coverage, reduce overtime and meet customer demand.” CPQ Test-to-Production Setup & Configuration Migration “analyzes a CPQ migration package and simulates its effects on affected production data.” Executive Briefing Document, the one new name, is not on the Help Center page, and its release describes a briefing “that it can update as account data changes.”
Oracle’s 26D documentation describes two of the five more narrowly than the releases. Account Territory Growth Plan’s page lists its actions as to “create opportunities, add accounts to engineered sales campaign programs, and create follow-up tasks for at-risk accounts,” for sales managers and sales VPs. Strategic Field Demand Planning’s page says “Recommendations are advisory,” and asks users to “Review any stated tradeoff, evidence status, effective date, or approval boundary before taking an operational action.” Greyhound Research found no 26D page for the other three.
“Business leaders need to turn growth strategies into opportunities and deliver on customer commitments, but connecting customer demand with enterprise capacity in time to act remains a challenge,” said Raghavan.
What follows automatically. In the release wording, four of the five carry approved or authorized action into execution. Wherever a change reaches the Fusion modules beneath them, made by the application or by a person acting on its advice, those modules pass it on by themselves.
Account Territory Growth Plan’s documented actions stop short of territories. When a sales leader acts on its coverage findings by changing territories, Fusion Sales makes the change through a proposal: “You can make changes to many territories at one time using a territory proposal,” Oracle’s documentation says. Oracle advises “Check your imported territories before you activate your proposal,” in its guide to proposals, and once territories are realigned, “Automatic assignments are always enabled following an import, party merge, or territory realignment.” One activated plan reassigns accounts across the sales force. Greyhound Research found no Oracle document describing how an activated proposal is reverted.
Service Health Check’s fixes, which span “knowledge, automation, routing, policy, and process,” reach customers through Fusion Service’s own mechanisms. Where queue assignment on creation is switched on, assignment rules run “when the SR is created,” in Oracle’s description, and by default, with a review workflow, “articles are published automatically after the workflow completes,” Oracle notes. An approved change to routing or knowledge reaches the next customer who calls or reads.
Strategic Field Demand Planning advises, and the changes made on its advice reach the field. In capacity-based quota management, Oracle Field Service defines quota as a percentage of capacity, and “This means that your quota updates automatically when the workforce changes,” Oracle’s documentation says. In Greyhound Research’s reading, a resource change made on the plan’s advice can change what customers are able to book.
The CPQ application addresses the one cascade with no documented way back. Oracle CPQ’s migration programming interface lists operations to “Version a Migration Package,” export, import and “Migrate Package,” in Oracle’s reference, and Greyhound Research found no rollback or restore for a migration in the Oracle documentation reviewed. The Claw application is built for the step before: “Administrators can find and resolve risks while the configuration is still in test,” the Help Center says. Greyhound Research credits that as the right control in the right place, because after migration every quote runs on the new configuration.
No customer spoke on the record about a Claw customer experience application at AI World, in any source Greyhound Research reviewed. Oracle’s own support organization offers the nearest signal. “What my support team spends all their time doing is how can we get more questions right, and how fast can we turn it around?” Leone said, describing how its work has changed.
Exhibit 7 sets out each customer experience application by what follows automatically, the way back found, and the checkpoint that belongs with it.
Greyhound Research · Exhibit 7 · Customer experience cascades
The Fusion modules beneath customer experience pass a change on by themselves, so the checkpoint belongs before the change
What the Fusion module does with a change, in Oracle’s words, the way back found, and the checkpoint proposed.
Put the checkpoint before the cascade, and keep the prior state before every change.
| Application | What follows automatically | Way back found | Checkpoint |
|---|---|---|---|
| Account Territory Growth PlanChanges: Opportunities, campaigns, tasks (26D) | Where territories are realigned on its findings, “Automatic assignments are always enabled” | No territory revert found | CRO approves any territory activation, prior territories kept |
| Service Health CheckChanges: Routing, knowledge, policy | Where switched on, queue rules run “when the SR is created,” and articles “are published automatically” by default | Change it again | Head of service approves each change |
| Strategic Field Demand PlanningChanges: Advice on field resources (26D) | A resource change made on its advice: “Your quota updates automatically” | Change it again | Field service lead approves each resource change |
| CPQ Test-to-Production Setup & Configuration MigrationChanges: Production configuration | Every quote runs on the migrated configuration | No rollback found | Keep the last configuration as a versioned package |
| Executive Briefing DocumentChanges: A briefing | Nothing, until a person uses it | Edit it | Account executive |
Exhibit 7 sets out the five customer experience applications in Oracle’s AI World releases. Account Territory Growth Plan’s 26D actions are opportunities, campaign enrollments and follow-up tasks, and where a sales leader realigns territories on its findings, automatic assignment reassigns accounts, with no documented revert found. Service Health Check changes routing, knowledge and policy, and, where switched on, queue rules run on new service requests while reviewed knowledge articles publish automatically by default. Strategic Field Demand Planning is documented for 26D as advisory, and a resource change made on its advice updates Field Service quota automatically. CPQ Test-to-Production Setup & Configuration Migration migrates configuration to production, where every quote runs on it, and no rollback or restore was found, so the Claw simulation in test is the control. Executive Briefing Document produces an editable briefing. For each, the exhibit names the checkpoint Greyhound Research proposes. Greyhound Research analysis of Oracle’s releases and documentation.
Greyhound Standpoint: Greyhound Research judges the customer experience applications to be well aimed and the least proven. They address real friction, demand that outruns capacity and service problems that grow before anyone sees them, and the CPQ application puts its check exactly where Oracle’s migration tooling has no way back. Oracle’s documentation describes the two with 26D pages more narrowly than their releases, and no customer has yet spoken on the record about running one.
Because the modules beneath them cascade, the checkpoint has to sit before the cascade. Chief revenue officers should approve any territory activation made on the growth plan’s findings themselves, with a record of the prior territories kept outside the proposal, and review the accounts it adds to sales campaign programs. Field service leads should treat the demand plan as the advice Oracle documents it to be, and approve each resource change made on it. Heads of service should treat a routing or knowledge change as a customer-facing change and approve each one until receipts show the pattern. CPQ owners should keep the last production configuration as a versioned package before every migration, so that the way back exists even if the tooling does not provide one. Executive Briefing Document, which produces a document a person edits, is the natural place to start.
The Handoffs: One Decision, Several Owners
The hardest ownership questions do not sit inside any one function. They sit where one application’s approved action becomes another function’s input. Greyhound Research traces four such chains through the AI World releases and the Fusion modules beneath them.
A field schedule becomes a payroll cost. A resource change made on Strategic Field Demand Planning’s advice, such as the overtime it can suggest, changes the hours technicians work. Where those hours include overtime, they reach payroll as time worked, payroll costs them, and once costing runs in final mode, Oracle’s documentation says, the transfer to the general ledger cannot be rolled back. One approval, by the field service lead, ends in an entry the controller books, with the head of payroll in between.
A workforce plan becomes a payslip. Workforce Orchestration recommends actions across development, redeployment, hiring and external talent, and in 26D a person runs them, since its workspace “doesn’t perform HCM transactions.” A position change flows to every synchronized assignment, a hire moves through recruiting to an approved offer, and the new employee’s first pay runs through payroll to the ledger. Four owners touch it: the business leader who approved the plan, HR operations, talent acquisition, and payroll, before the controller books the cost.
An allocation becomes a delivery promise. An allocation approved in Supply Planning Allocation decides which customers receive scarce stock. The loads that carry it are planned through Shipment Load Planning and tendered to carriers, and the customer left short hears from sales. The planner, the logistics lead and the account owner each hold a piece of one decision.
A configuration becomes every quote. A migration approved through the CPQ application changes the configuration every quote runs on. Sellers quote on it, orders follow, and revenue is booked on the prices it produced. The CPQ owner approves the migration, and the controller inherits it.
Each application’s outcome record describes its own run. The research report of October found the record’s export, retention and link to Fusion’s audit tables not yet documented, and no Oracle document reviewed for this note describes linking one application’s outcome record to the next application’s. Oracle’s technical brief lists “linked transactions” among what a receipt records, which ties a receipt to the transactions of its own run, not to the receipt of the next application. Across a chain, the account of what happened is reconstructed by people. Mapped in Exhibit 8, each chain crosses at least three owners.
Greyhound Research · Exhibit 8 · The handoffs
One approved action can cross three to five owners before it reaches the ledger, and no outcome record yet links the steps
Four chains in which one Claw application’s output becomes another function’s input, with the owner at each step.
Give every chain an owner, and let each owner downstream see the approval upstream.
Exhibit 8 traces four chains in which one Claw application’s output becomes another function’s input. A field schedule: a resource change made on Strategic Field Demand Planning’s advice changes technicians’ hours, payroll pays and costs them, and costing transferred to the ledger in final mode cannot be rolled back, crossing the field service lead, the head of payroll and the controller. A workforce plan: Workforce Orchestration recommends actions that a person runs, position changes flow to synchronized assignments, recruiting moves a hire to an approved offer, and payroll pays, crossing the business leader, HR operations, talent acquisition, payroll and the controller. An allocation: Supply Planning Allocation decides who receives scarce stock, Shipment Load Planning plans the loads tendered to carriers, and sales tells the customer left short. A configuration: the CPQ application migrates a configuration, every quote runs on it, and revenue is booked on its prices. For each, the exhibit marks where an approval sits and where no Oracle document describes a link between outcome records. Greyhound Research analysis.
Greyhound Standpoint: Greyhound Research holds that the handoffs are where ownership by function will be tested first. Each function head can own their application well and still leave a gap at the point where an approved schedule becomes someone else’s payroll cost, or an approved migration becomes someone else’s revenue. The receipt that makes one application auditable does not yet make a chain auditable.
The fix is an owner for each chain as well as for each application. CIOs should keep a register of the chains their Claw estate creates, with the owner at each step, the point where each step is approved, and the record that carries the decision forward. Each owner downstream should be able to see the approval upstream before acting on it, and the controller should be able to trace an entry back through the chain to the first approval.
One Runtime, Four Owners: Budgets, Stops, And The People Around Them
Across the four functions, the runtime is one and the owners are many. Oracle has built the controls that cross functions into the runtime and its documentation. The owners are the buyer’s to name.
The economics are designed to be held application by application. “We only use AI units when we’re reasoning,” Leone said in a customer session at the event. At the Fusion Applications press conference, he said customers can “set budgets at a macro-level,” and at “an agent level, and an agentic application level,” and that “We have a kill switch.”
Oracle’s documentation carries the same design. Administrators can “assign budgets to individual applications or agents,” and the page lists “Configure warning thresholds and notification recipients” and “Enable auto-stop to prevent new runs when a selected budget is exhausted” among its controls. Administrators can also stop or resume all items, in Oracle’s words “when immediate operational control is required, including options to stop only new sessions or also stop all running sessions,” and “Alert and change histories provide an auditable record of budget changes, threshold events, and control actions.” Leone was candid about the limit and about Oracle’s answer to it. “It’s really hard to predict what the cost is going to be at the upfront,” he said, but Oracle “will go run a simulation to tell you what the estimate will be,” and “create an estimated budget that you can then set in the application.”
Oracle’s service terms add a stop the buyer does not hold. “If Your available Fusion AI Units fall below zero, Oracle may limit, reduce, or block access to some or all functionality that consumes AI Units,” the Fusion Service Descriptions say, and units “that remain unused or unapplied at the end of the Services Period will be forfeited.”
Greyhound Research reads the budget design as the piece that maps most directly onto ownership by function. A budget per application lets the controller hold the budget for finance, the CHRO the budget for the HR applications, the chief supply chain officer the budget for theirs, the chief revenue officer the budget for customer experience, and the CIO the total. Warning thresholds can go to the owner who should act on them. The two stops divide the same way. Auto-stop halts new runs of one application when its budget is spent, and belongs with its owner. The stop for “all items” halts the fleet, and belongs with the CIO. Oracle’s own stop, when units run out, makes the total a number the CIO and the CFO watch together.
The operating model is the harder part, and Oracle’s leaders say so. “Most of the customers are at least six months behind,” said Rondy Ng, executive vice president, Applications Development, at Oracle. “Most of our existing customers are still deploying the agents that we wrote six, nine months ago,” he said.
Asked how system integrators had reacted to Oracle making Fusion easier to configure, Leone said that “roles will have to evolve into managing the governance layer of these agents, and managing ultimately the standard operating procedures that they operate under.” Gary Miller, executive vice president and customer success officer at Oracle, said “We’re starting to see some emerging patterns, but it’s really early,” and described the question he has put to his own HR lead: “my team now looks like humans and agents, what does the org chart look like?” Miller added, in diginomica’s account, that Oracle has defined “approximately 100 ‘value paths’ for Fusion,” setting out the human and agent tasks in processes such as recruiting.
Customers named the same pairing. “Have a duo between business and IT – shared responsibility is a big key,” said Passarge of Deutsche Bahn. That is the ownership model in one line. The function owns the outcome, IT owns the platform, and both sign. Table 4 sets out the ownership map Greyhound Research proposes, including the handoffs each function shares. In it, the CHRO signs with the head of payroll and the head of talent acquisition, the chief supply chain officer with the planning, logistics and procurement leads, and the chief revenue officer with the head of service, while each function’s application budget sits with the function head alone.
Greyhound Research · Table 4 · Ownership map
Four functions, four owners, one runtime, and the shared handoffs
Who signs and holds the budget, who co-signs, and the handoffs each function shares. Greyhound Research’s proposal.
The function owns the outcome, IT owns the platform, and both sign.
| Function | Signs, and holds the budget | Co-signs | Handoffs it shares |
|---|---|---|---|
| Finance | Controller | CFOfor amounts above a set value | Receives every chain’s postings |
| HR | CHROwith the head of payroll and the head of talent acquisition | CFOfor pay and payrollGeneral counselfor hiring and outreach | Plans to positions, hiring, payrollField hours into payroll |
| Supply chain | Chief supply chain officerwith the planning, logistics, and procurement leads | Head of salesfor allocations that leave a customer shortCFOfor spend above a set value | Allocations to loads to customersSpend to suppliers and payables |
| Customer experience | Chief revenue officerwith the head of service | Pricing leadfor configuration quotes run onHead of payrollfor field schedule changes | Field schedules into payrollConfiguration into quotes, revenue |
| Across all four | CIO | CFOfor the total budget | Chain register, fleet stop, one authority model |
Table 4 sets out the ownership map Greyhound Research proposes for each function: the owner who signs and holds the application budget, the co-signers, and the handoffs the function shares. The controller owns Ledger, with the CFO above a set value, and receives the postings at the end of every chain. The CHRO owns the HR applications with the head of payroll and the head of talent acquisition, with the CFO co-signing pay and payroll and general counsel co-signing hiring and outreach. The chief supply chain officer owns the supply chain applications, with the head of sales co-signing allocations that leave a customer short and the CFO co-signing spend above a set value. The chief revenue officer and the head of service own the customer experience applications, with the pricing lead and the head of payroll as co-signers. Across all four, the CIO holds the total budget with the CFO, the chain register, the fleet stop, and one authority model. Greyhound Research analysis, current through October 30, 2026.
Greyhound Standpoint: Greyhound Research takes the view that Oracle’s cross-function controls suit ownership by function well. Budgets per application, thresholds sent to named recipients, a stop when a budget is spent and an auditable history of budget actions are the instruments a function owner needs, and they are documented. Real-time cost monitoring is, in diginomica’s account of Leone’s remarks, a work in progress. Oracle’s own leaders have been candid that the operating model, not the software, is where customers are behind.
The work is to give every instrument a person. CIOs should assign each application’s budget and warning threshold to the function owner in Table 4, keep the fleet-wide stop, the total budget and the chain register themselves, and agree with the CFO a single total for AI units across the four functions, with the point at which Oracle may limit access written into the plan. A budget no one in the function watches becomes a cost the CIO explains later.
The Calendar: Release 26D Meets Each Function’s Year
The applications arrive by calendar, and the calendar is not neutral. Oracle’s quarterly updates reach test environments “on the first Friday of the update month” and production “two weeks later, on the third Friday,” Oracle’s Fusion team writes, in three cohorts. Oracle’s readiness pages for ERP, HCM, supply chain, sales and service each say “Explore the new Fusion Claw capabilities available in update 26D!” Applying the cohort rule to Release 26D, as Greyhound Research reads Oracle’s update naming, puts production on November 20, 2026, December 18, 2026, or January 15, 2027, depending on cohort.
Arrival is not the same as running. The quarterly updates “are mandatory for all customers,” Oracle’s environment documentation says, but new features that materially change the user experience “are shipped ‘disabled’ and require opt-in,” the Fusion team writes, and every Claw application documented so far needs a setup step, a role, a profile option, a policy or a configured outcome, before it runs. The cohort is fixed once chosen: “After you create an environment family, you can enable or disable monthly patching. You can’t change the quarterly maintenance schedule,” the same documentation says.
Payroll customers do not choose. “Oracle Fusion Cloud Payroll has mandated maintenance periods in February/May/August/November,” Oracle’s documentation says, and “you can’t change the maintenance months.” Every environment family that runs Oracle payroll is in the first cohort. If the payroll applications ship in update 26D, which Oracle’s HCM 26D guides do not yet show, since neither payroll application has a 26D page, they arrive in those customers’ test environments on November 6 and production on November 20, about six weeks before year-end for a calendar-year payroll.
Where the payroll year ends elsewhere in the calendar, as in Australia, where the financial year “starts on 1 July and ends on the next 30 June,” the Australian government’s Study Australia guide notes, the collision moves, and the planning does not.
The other functions meet the update at different points in their year. For calendar-year companies, the year-end close begins in the weeks after the second and third cohort dates, which is when the controller is least likely to want a new application acting on journals. Employers whose annual pay review runs in the first quarter could run their first Claw-assisted cycle weeks after the update, if Workforce Compensation ships in 26D, before the first pay transparency reports fall due.
For retailers and consumer goods makers, the first two cohort dates fall in peak shipping season. For sellers on a calendar fiscal year, new-year territory and pipeline plans are set in exactly these weeks, in Greyhound Research’s reading, which makes the timing right for Account Territory Growth Plan’s pipeline and campaign work.
The rules keep their own dates. Illinois is in force, Colorado and California apply from January 1, 2027, the first pay transparency reports are due by June 7, 2027, and the EU AI Act’s high-risk obligations follow on December 2, 2027. Laid out in Exhibit 9, the update lands well for one function and awkwardly for the rest.
Greyhound Research · Exhibit 9 · The calendar
Release 26D arrives well timed for sales planning, and in the busiest weeks of the year for payroll, finance, and supply chain
Release 26D’s three production dates, from Oracle’s cohort rule, against each function’s year and the rule dates.
Let each function’s year set its start, not the update date. Payroll cannot choose a later cohort.
Exhibit 9 lays the three production dates of Release 26D, derived from Oracle’s published cohort rule, against each function’s year from November 2026 to June 2027. Cohort A reaches test on November 6 and production on November 20, 2026, Cohort B production on December 18, 2026, and Cohort C production on January 15, 2027. Payroll is locked to Cohort A by Oracle’s mandated maintenance months and meets calendar year-end six weeks later. Finance meets the calendar-year close in January. Compensation meets annual reviews in the first quarter for many employers. Supply chain meets peak season in November and December for retailers and consumer goods makers. Sales sets new-year plans in the same weeks, which suits Account Territory Growth Plan. A lower band marks rule dates: Illinois in force, Colorado and California from January 1, 2027, and the first EU pay transparency reports by June 7, 2027, and a note records that the EU AI Act’s high-risk obligations apply from December 2, 2027, beyond the window. The update arrives on these dates, and each application still needs a setup step before it runs. Greyhound Research analysis of Oracle sources and the rules cited.
Greyhound Standpoint: Greyhound Research regards the calendar as the most practical finding in the AI World releases for most buyers, because it decides when each function can first switch Claw on. Oracle’s cohort rule is published and predictable, the payroll lock is documented, and every application documented so far needs a setup step before it runs, which lets a buyer plan now rather than discover the timing in November.
The answer is to let each function’s year set its start, not the update date. Heads of payroll should run Payroll Command Center’s checks, when the application reaches them, beside the existing year-end validation, not instead of it, and act on its findings in the live cycle from the first payroll of the new year. Controllers should keep Ledger in analysis, with automatic corrections and scheduled runs off, through the year-end close. Supply chain should hold its first external commitments until after peak, and compensation owners should finish documenting their criteria before the first-quarter cycle opens. Sales can start at once.
Risks To Watch: Where Claw Across Functions Can Go Wrong
Claw fails differently in each function, because each function’s applications commit something different and each module beneath them comes back a different way. Six risks stand out, and between them they reach all four functions.
One identity can prepare and approve. Where an application runs under the approver’s own token, by the rule Oracle has published for AI Agent Studio, and the approver then accepts its recommendation, the receipt would show one person as both requester and approver. Watch for approvals given by the user who ran the outcome, and for agent or service-account roles that no segregation of duties rule covers. The remedy is to bar self-approval in the processes that move money, and bring agent and service-account roles into the same rules as people, owned by the controller with the head of internal audit.
Year-end payroll meets a new application. Payroll customers take the update in the first cohort, weeks before a calendar-year close, and if the payroll applications arrive in 26D, they arrive with it. Watch for findings acted on after prepayments, and for corrections made by rollback, which in Oracle’s documentation leave no audit trail in payroll. The remedy is to run the checks beside the existing year-end validation, act before prepayments, and record every rollback decision outside payroll, owned by the head of payroll.
A message sent cannot be recalled. Candidate Sourcing coordinates outreach and follow-up to candidates, and its release does not say that a person presses Send. Watch for campaigns launched before legal has mapped them, for replies an AI agent sends without a person, or for New York City roles without a bias audit and notice. The remedy is per-message approval until the mapping is done, with audit and notice where the law requires them, owned by the head of talent acquisition with general counsel.
One approval can change many records. A position change flows to every synchronized assignment, a territory realignment made on the growth plan’s findings reassigns accounts automatically, and a field resource change made on the demand plan’s advice updates quota by itself. Watch for approvals given without a count of the records they will change. The remedy is a second approver above a set number of records affected, and the prior state saved before each change, owned by the HR operations lead, the chief revenue officer and the field service lead.
The way back can cost money, or not exist. A tender withdrawn after acceptance can re-tender at higher cost, the change-order statuses Oracle names for withdrawal are all pending ones, and Greyhound Research found no documented rollback for a CPQ migration. Watch for external commitments released without a named planner, and migrations made without the last configuration saved. The remedy is a planner or buyer who releases each external commitment, and a versioned package of the prior configuration before every migration, owned by the chief supply chain officer and the CPQ owner.
Cost and chains can scatter across four owners. Budgets per application let each owner hold their own, and nobody may hold the total or the chain. Watch for one function reaching auto-stop while the total rises unseen, for units running short of the point at which Oracle may limit access, and for a chain whose steps no one can trace end to end. The remedy is one total budget held by the CIO with the CFO and a register of chains with an owner at each step, owned by the CIO.
Placed by function in Exhibit 10, every risk has a remedy that sits with a role the buyer already has, and the early warnings above tell each owner what to watch.
Greyhound Research · Exhibit 10 · Risk map
Six risks reach every function, and each remedy has an existing owner
Six risks of Claw, the functions each reaches, its remedy, and its owner. The early warnings are in the text.
Count how many approved actions a person later changed, reversed, or rolled back. It shows whether approval is a check or a formality.
| Risk | Finance | HR | Supply chain | Customer experience | Remedy | Owner |
|---|---|---|---|---|---|---|
| Risk 1One identity can prepare and approve | Bar self-approval, and map agent roles into the same rules as people | Controller, with internal audit | ||||
| Risk 2Year-end payroll meets a new application | Run the checks beside year-end validation, and act before prepayments | Head of payroll | ||||
| Risk 3A message sent cannot be recalled | Approve each message until legal has mapped the outreach | Head of talent acquisition, with general counsel | ||||
| Risk 4One approval can change many records | A second approver above a set record count, with the prior state saved | HR operations lead, CRO, field service lead | ||||
| Risk 5The way back can cost money, or not exist | A named person releases every external commitment | Chief supply chain officer, CPQ owner | ||||
| Risk 6Cost and chains can scatter across four owners | One total budget, and a register of chains | CIO, with the CFO | ||||
| Risks reaching it | 2 | 5 | 3 | 3 |
Exhibit 10 lists six risks of Claw across functions, each tagged with the functions it reaches. One identity preparing and approving reaches finance and every function that moves money. Year-end payroll meeting a new application sits in HR. A message that cannot be recalled sits in HR recruiting. One approval changing many records reaches HR positions, customer experience territories and field service. A way back that costs money or does not exist reaches supply chain tenders and supplier change orders and the CPQ migration in customer experience. Cost and chains scattering across four owners reaches all four functions. For each, the exhibit gives the remedy and its owner, and counts how many risks reach each function: five of the six reach HR. One early warning serves all six: how many approved actions a person later changed, reversed or rolled back. Greyhound Research analysis.
Greyhound Standpoint: Greyhound Research’s assessment is that all six risks are manageable, and none argues against switching Claw on in any function. Each argues for an owner who reads what the application did, not only what it proposed, and who knows how the module beneath it comes back. The risks share one cause: an approval designed for one decision, used for many, or for someone else’s.
CIOs should give each risk an owner and an early-warning measure, and ask function heads to report them beside the gains. Oracle’s own product leadership allows that reading may fade. “After a point, maybe you won’t look at every single outcome receipt,” Kaushal Kurapati, GVP Product Management, Oracle Fusion AI, told CFOtech. “Maybe you will let go.” He also said “there are some decisions that are obviously taken by the human because accountability is at stake.” The measure that matters most is the same in every function: how many approved actions a person later changed, reversed or rolled back. That number shows whether approval is a check or a formality, and whether letting go has been earned.
The Greyhound CXO Runbook: Six Moves, One For Each Leader
Six moves follow from the evidence, and they are written for the leaders who will each own part of the Claw estate. Each move names who decides. None waits for Oracle, because owners, signatures, approvers, budgets and the order of adoption are already the buyer’s to set. For the setting each process should start at, The Delegation Dial sets out the choices process by process.
1/ Name an owner for every application and every chain before your cohort’s update arrives.
Take Table 1 and Exhibit 2, and give every application the organization plans to use a named owner, a deputy, and a named approver in place of Oracle’s “teams,” “recruiter” and “organizational review.” Find each application by name in your cohort’s update and in Oracle’s 26D readiness pages, as in Table 2, check its default setting before it is switched on, and record the chains in Exhibit 8 that your estate will create, with an owner at each step.
An application or a chain without an owner should not be switched on, however ready it is. The decision sits with the CIO, together with each function head.
2/ CFOs: set the signature rule, then map the money.
Bar self-approval of Claw recommendations in payroll, procurement and the close, run scheduled outcomes only under service accounts whose privileges are kept apart from approvers’, and bring agent and service-account roles into the same segregation of duties rules as people, through Advanced Access Controls or an equivalent.
Then list the applications that name money, as in Exhibit 3, and agree where finance co-signs: every compensation plan, every payroll correction, and corrective spend above a set value. Keep Ledger in analysis, with automatic corrections and scheduled runs off, through the year-end close. Before Collector Workspace’s Claw capabilities become available, which Oracle dates “starting 13th Nov CWB,” approve the collections policy it executes and name the administrator under whose user context it reads the collections mailbox. The decision sits with the CFO, together with the controller and the head of internal audit.
3/ CHROs: start with the checks, and map the law before the decisions.
When they reach the estate, run Payroll Command Center and Payroll Activity Workspace beside the existing year-end validation, act on findings before prepayments, and require a second approver for position changes that reach many assignments. Before Workforce Compensation, Recruiting Command Center or Candidate Sourcing runs, have general counsel map each to the rules in Table 3 for every location the employer hires or pays in.
Keep Candidate Sourcing on per-message approval, replies included, require each Workforce Compensation recommendation to record the criteria it weighed, and inform worker representatives before the first run. The decision sits with the CHRO, together with general counsel.
4/ Chief supply chain officers: go first, after peak, and keep the commitment with a planner.
Hand over the analysis from the first cohort, and hold the first external commitments until after peak season. Keep the release of tenders, allocations that leave a customer short, and supplier change orders with a named planner or buyer, because each way back narrows or costs more once the other party has the commitment.
Bring sales into any allocation that takes supply from one customer for another. The decision sits with the chief supply chain officer, together with the procurement and logistics leads.
5/ Chief revenue officers and heads of service: put the checkpoint before the cascade.
Use Account Territory Growth Plan for the new year’s pipeline and campaign work, approve any territory activation made on its findings yourself, and keep a record of the prior territories. Have field service leads approve each resource change made on Strategic Field Demand Planning’s advice. Treat every routing and knowledge change from Service Health Check as a customer-facing change, approved one at a time until receipts show the pattern. Before every CPQ migration, keep the last production configuration as a versioned package.
Start with Executive Briefing Document, which produces a document a person edits. The decision sits with the chief revenue officer, together with the head of service.
6/ CIOs: keep one authority model, one total budget, one stop, and a calendar set by function.
Assign each application’s budget and warning threshold to its owner, with auto-stop on, and hold the total budget for AI units, the fleet-wide stop and the chain register yourself, with the point at which Oracle may limit access written into the total. Keep one authority model across the four functions, so that an approval means the same thing in HR as in customer experience, and set each function’s first month by its own year, as in Exhibit 9.
Report the number of approved actions later changed, reversed or rolled back, by function, to the leadership team each quarter. The decision sits with the CIO, together with the CFO.
Sequenced in Exhibit 11 against the update calendar and each function’s year, four of the six moves start now.
Greyhound Research · Exhibit 11 · Runbook sequence
Four Runbook moves start now, payroll’s checks run beside its year-end, and HR’s decision applications follow the legal map
The six Runbook moves against the update calendar and each function’s year, with the owner of each decision.
Owners, signatures, and the legal map come before the first run. No move waits for Oracle.
Exhibit 11 sequences the six Runbook moves across five phases: now, before the test update, from production to the year-end close, the first quarter of 2027, and ongoing. Naming owners and chains, the CFO’s signature rule, the CHRO’s legal mapping and criteria, and the CIO’s budgets, stop and calendar start now. The payroll checks run beside year-end validation from the test update through the close. Sales planning for the new year starts at the production update. Supply chain hands over analysis from production and releases its first external commitments after peak, in the first quarter. HR’s decision applications follow the legal mapping in the first quarter. Each bar is labeled with its owner. Release 26D dates are derived from Oracle’s published cohort rule, and payroll environments take the first cohort.
Questions For Oracle: What Buyers Will Ask Next
Q1 Buyers plan on names and dates. Oracle says twenty-five Claw-powered applications are available, its AI World releases name sixteen, its Help Center page lists fifteen capabilities, and its documents together carry twenty-four distinct names. None of the seven new names appears on the Help Center page, and none of the six new HR names has a 26D page that describes it as Claw-powered yet. Which twenty-five applications are Claw-powered, by name, which reach customers in each cohort’s update, including the six new HR applications that payroll customers would receive in the first cohort, and will Oracle publish for each the value path that sets out what people and agents do?
Q2 Auditors will ask who prepared and who approved. Oracle’s technical brief says the outcome receipt “records the requester, owner, executor, and approver,” and its AI Agent Studio guidance says “Every action carries the running user’s Fusion token.” Where the approver also ran the outcome, the preparer and the approver can be the same identity. Which identity does a Claw run execute under, how does a receipt map to Fusion’s audit records, how long are receipts kept, given the EU AI Act’s six-month minimum for high-risk system logs, and will Advanced Access Controls analyze agent and service-account roles natively, rather than as a reference pattern?
Q3 HR buyers will ask what a pay recommendation rests on. Workforce Compensation evaluates options against “employer-defined criteria, budgets, pay guidelines, and relevant equity considerations,” and the EU Pay Transparency Directive asks for “objective, gender-neutral criteria.” What are the equity considerations, can an employer see, change and test them, and does each recommendation record the criteria it applied, so that the record can support a pay transparency report?
Q4 Hiring law is already enforced. Oracle’s own Candidate Outreach feature does not send AI-written emails “until you select Send,” though its AI agent can answer a candidate’s reply. Does Candidate Sourcing send outreach or campaigns without a person selecting Send, and what documentation does Oracle provide to support a bias audit under New York City’s law and an employer’s duties under the EU AI Act?
Q5 Every action needs a way back. In Oracle’s payroll documentation, a rollback leaves “no audit trail,” and Greyhound Research found no documented revert for an activated territory proposal or a CPQ migration. Does a Claw outcome’s record survive a payroll rollback, and will Oracle document how territory activations made on Claw’s findings, and CPQ migrations made through Claw, are reversed?
Q6 Decisions cross functions. A field schedule becomes a payroll cost, and a configuration becomes every quote. Will a Claw outcome’s record link to the outcome records and transactions it causes in other applications, so that a controller can trace a ledger entry back to the first approval?
Q7 Technology leaders are being asked to account for AI, not only adopt it. Greyhound CIO Pulse 2026, across more than 1,000 technology leaders, finds fewer than one in three can point to measurable outcomes across most of their funded AI portfolio, and Ng has said most customers are “at least six months behind.” Oracle’s service descriptions meter AI units by type of action, and Leone has described a simulation that estimates a budget, but Oracle’s releases and documentation reviewed name no customer running a Claw application in production, and state no units per outcome. Which customers run a Claw application in production in each function, and what units does a typical outcome consume for one application in each?
Read together in Table 5, each of the seven questions concerns something Oracle has named but not yet written down in full, and each maps to a Runbook move a buyer can make meanwhile.
Greyhound Research · Table 5 · Questions tracker
Each question concerns something Oracle has named but not yet written down in full, and each comes with a move for the buyer
Each question, the buyer signal behind it, its status in Oracle’s documents, and the Runbook move meanwhile.
Until Oracle answers, each question maps to a Runbook move the buyer controls.
| Question for Oracle | Buyer signal | In Oracle’s documents | Move |
|---|---|---|---|
| Q1The twenty-five by name, and each one’s cohort | Names and dates before plans | 24 names, no single list | 1/6/ |
| Q2Which identity a Claw run executes under, and segregation of duties for agent roles | Preparer and approver kept apart | Receipt fields documented, run identity not | 2/ |
| Q3Workforce Compensation’s equity criteria, recorded per recommendation | Pay decisions the employer can explain | Criteria named, not described | 3/ |
| Q4Whether Candidate Sourcing sends without a person, and audit support | Hiring law already enforced | Not described for the Claw application | 3/ |
| Q5Outcome records after a rollback, and reverts for territories and CPQ | A way back for every action | Not described | 3/5/ |
| Q6Outcome records linked across applications | Every entry traceable to its approval | Not described | 1/6/ |
| Q7Named production customers, and units per outcome | Peers and costs on the record | No production customer named, no units stated | 4/6/ |
Table 5 maps the seven questions for Oracle to the buyer signal behind each, the status in Oracle’s documents, and the Runbook move a buyer can make meanwhile. The twenty-five applications by name, the identity a Claw run executes under, where the receipt’s fields are documented but the run identity is not, and Workforce Compensation’s equity considerations are named or described in part. Whether Candidate Sourcing sends without a person, outcome records after a payroll rollback and reverts for territories and CPQ migrations, outcome records linked across applications, and named production customers by function with units per outcome are not described. Current through October 30, 2026.
Fusion Claw By Function: The Final Greyhound Standpoint
Greyhound Research’s final judgment is that Oracle has taken Fusion Claw into every function and has described each application, in its releases, with the point at which a person comes in. The sixteen applications in its AI World releases span finance, HR, supply chain and customer experience. Every release description ends with a person, five of the seven HR applications stop at a recommendation, and the HR release writes employer responsibility into Oracle’s own words. Oracle’s 26D documentation, where it exists, is more cautious than the releases in HR and customer experience, and more autonomous in finance.
Read from the module up, the decision is more granular than the releases suggest. Each application lands on a Fusion process with its own approvals and its own way back, and those ways back differ: a journal reversed once, in the same or a later period, a payroll rolled back without an audit trail, a tender withdrawn at a higher price, a CPQ migration with no documented rollback.
In HR, the law names the work the decision applications do, and in every function that moves money, the receipt can show one person as both requester and approver unless someone decides otherwise. The update itself arrives in the busiest weeks of the year for payroll, finance and supply chain, and every application Oracle has documented so far needs a setup step before it runs.
The position would strengthen with one list of Claw-powered applications by name for each cohort, documentation of the identity each Claw run executes under and of how its receipt maps to Fusion’s audit records, with native segregation of duties analysis of agent roles, criteria recorded with each pay recommendation, per-message approval documented for Candidate Sourcing, outcome records linked across applications, and a named customer running a Claw application in production in each function. It would weaken if self-approved outcomes appeared in the processes that move money, if HR decisions came to be approved in bulk, if cascading changes went live without a checkpoint, or if costs rose across functions with no one holding the total.
Of the ten signals in Exhibit 12, four would show up first in a buyer’s own estate rather than in anything Oracle publishes.
Until then, the buyers who gain most will be those who treat ownership as the control, down to the module. They will name an owner for every application and every chain, check each application’s 26D page and default before it runs, keep the preparer and the approver apart, start HR with the checks and the legal map, and know how each module beneath an application comes back before they let it act. They will let supply chain go first after peak, put the checkpoint before every cascade, and set each function’s start by its own year rather than the update date.
Leone described software that does the work “while people define the objectives and set the policies.” Across four functions, those people now have to be named, and the signature each of them gives has to be their own.
Greyhound Research · Exhibit 12 · Signals to watch
The position holds, and ten signals would move it
What would strengthen or weaken Greyhound Research’s position on Fusion Claw, and where a buyer can watch for each.
Four of the ten signals show up first in the buyer’s own estate, not in anything Oracle publishes.
Current position: holds. Breadth delivered, with a person at every application. Signatures, chains, and evidence to come.
Would weaken the position
Would strengthen the position
Exhibit 12 lists ten signals that would move Greyhound Research’s position on Fusion Claw across functions. Six would strengthen it: one list of Claw-powered applications by name for each cohort, documentation of the identity each Claw run executes under and of how its receipt maps to Fusion’s audit records, with native segregation of duties analysis of agent roles, criteria recorded with each Workforce Compensation recommendation, per-message approval documented for Candidate Sourcing, outcome records linked across applications, and a named customer in production in each function. Four would weaken it, and would show up first in the buyer’s own estate: self-approved outcomes in the processes that move money, HR decisions approved in bulk, cascading changes live without a checkpoint, and costs rising across functions with no one holding the total. Each signal names where a buyer can watch for it.
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. Everything in the first four classes is Oracle’s own account of itself and proves what Oracle said rather than that the thing is true. Standing pages that carry no publication date are marked n.d., with the release level shown on the page where there is one. All sources were accessed on October 29 or 30, 2026.
Oracle announcements
Oracle Corporation (2026a) Oracle Introduces Claw-Powered Fusion Agentic Applications for HR. Press release, October 21. Available at: https://www.oracle.com/news/announcement/oracle-introduces-claw-powered-fusion-agentic-applications-for-hr-2026-10-21/
Oracle Corporation (2026b) Oracle Advances Outcome-Driven AI Across Fusion Applications. Press release, October 26. Available at: https://www.oracle.com/news/announcement/ai-world-oracle-advances-outcome-driven-ai-across-fusion-applications-2026-10-26/
Oracle Corporation (2026c) Oracle Introduces Claw-Powered Fusion Agentic Applications for Supply Chain. Press release, October 26. Available at: https://www.oracle.com/news/announcement/ai-world-oracle-introduces-claw-powered-fusion-agentic-applications-for-supply-chain-2026-10-26/
Oracle Corporation (2026d) Oracle Introduces Claw-Powered Fusion Agentic Applications for Customer Experience. Press release, October 26. Available at: https://www.oracle.com/news/announcement/ai-world-oracle-introduces-claw-powered-fusion-agentic-applications-for-customer-experience-2026-10-26/
Oracle Corporation (2026e) Oracle Extends Fusion Agentic Applications with Introduction of Fusion Claw. Press release, September 29. Available at: https://www.oracle.com/news/announcement/oracle-extends-fusion-agentic-applications-with-introduction-of-fusion-claw-2026-09-29/
Oracle Corporation (2026f) Oracle Introduces Fusion Agentic Applications. Press release, March 24. Available at: https://www.oracle.com/news/announcement/oracle-introduces-fusion-agentic-applications-2026-03-24/
Oracle Corporation (2026g) Oracle Introduces Fusion Agentic Applications for HR. Press release, April 9. Available at: https://www.oracle.com/news/announcement/oracle-introduces-fusion-agentic-applications-for-hr-2026-04-09/
Oracle Corporation (2026h) Oracle Introduces Fusion Agentic Applications for Customer Experience. Press release, April 9. Available at: https://www.oracle.com/news/announcement/oracle-introduces-fusion-agentic-applications-for-cx-2026-04-09/
Oracle Corporation (2026i) Oracle Adds New Fusion Agentic Applications and AI Agents to Help Organizations Improve Talent Management. Press release, August 11. Available at: https://www.oracle.com/news/announcement/oracle-adds-new-fusion-agentic-applications-and-ai-agents-to-help-improve-talent-management-2026-08-11/
Oracle documentation
Oracle Cloud Infrastructure (n.d.) Planning an Environment Family. Fusion Applications environment management documentation. Available at: https://docs.oracle.com/en-us/iaas/Content/fusion-applications/plan-environment-family.htm
Oracle Corporation (2026k) Oracle Fusion Claw: The autonomous, governed execution runtime for Oracle Fusion Applications. Technical brief, Version 1.1, October. Available at: https://www.oracle.com/a/ocom/docs/oracle-fusion-claw-technical-brief.pdf
Oracle Corporation (2026l) Oracle Fusion Service Descriptions, v102026. Effective October 20. Available at: https://www.oracle.com/contracts/docs/oracle-fusion-cloud-service-desc-1843611.pdf
Oracle Help Center (2025) Create Purchase Order Changes as a Requester Using an Enhanced Redwood Change Order Page. Self Service Procurement, Update 25D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/scm/25d/ssproc25d/25D-ssproc-wn-f38825.htm
Oracle Help Center (2026a) Manage AI unit budget, consumption, and controls. Release 26D Common Technologies and User Experience What’s New, feature f51348. Available at: https://docs.oracle.com/en/cloud/saas/readiness/common/26d/common26d/26D-common-wn-f51348.htm
Oracle Help Center (2026b) Using General Ledger, Release 26B. Available at: https://docs.oracle.com/en/cloud/saas/financials/26b/faugl/using-general-ledger.pdf
Oracle Help Center (2026c) How to Prevent a General Ledger Period from Closing When Open Subledger Periods Exist, Release 26B. Available at: https://docs.oracle.com/en/cloud/saas/financials/26b/faugl/how-to-prevent-a-general-ledger-period-from-closing-when-open.html
Oracle Help Center (2026d) Overview of Workforce Compensation Plan Configuration. Available at: https://docs.oracle.com/en/cloud/saas/human-resources/faiwc/overview-of-workforce-compensation-plan-configuration.html
Oracle Help Center (2026e) How Workforce Compensation Plan Notifications Work. Available at: https://docs.oracle.com/en/cloud/saas/human-resources/faiwc/how-workforce-compensation-plan-notifications-work.html
Oracle Help Center (2026f) Other Details of Transfer Workforce Compensation Data to HR Process. Available at: https://docs.oracle.com/en/cloud/saas/human-resources/faiwc/other-details-of-transfer-workforce-compensation-data-to-hr-process.html
Oracle Help Center (2026g) Back Out and Purge Workforce Compensation Data Processes. Available at: https://docs.oracle.com/en/cloud/saas/human-resources/faiwc/back-out-and-purge-workforce-compensation-data-processes.html
Oracle Help Center (2026h) Corrective Actions for Payments. Administering Payroll for the United States. Available at: https://docs.oracle.com/en/cloud/saas/human-resources/faaus/corrective-actions-for-payments.html
Oracle Help Center (2026i) What’s the difference between Create Draft and Create Final Accounting for Payroll? Payroll costing documentation. Available at: https://docs.oracle.com/en/cloud/saas/human-resources/faipc/what-s-the-difference-between-create-draft-and-create-final.html
Oracle Help Center (2026j) Position Synchronization. Available at: https://docs.oracle.com/en/cloud/saas/human-resources/faucf/position-synchronization.html
Oracle Help Center (2026k) What are the phases and states of a candidate selection process? Available at: https://docs.oracle.com/en/cloud/saas/talent-management/farqa/what-are-the-phases-and-states-of-a-candidate-selection-process.html
Oracle Help Center (2026l) Job Offer Life Cycle. Available at: https://docs.oracle.com/en/cloud/saas/talent-management/faush/job-offer-life-cycle.html
Oracle Help Center (2026m) Redwood Experience: Candidate Outreach. Recruiting, Update 26C What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/hcm/26c/recr-26c/26C-recruiting-wn-f46428.htm
Oracle Help Center (2026n) Overview of Oracle Advanced Controls, Release 26A. Available at: https://docs.oracle.com/en/cloud/saas/risk-management-and-compliance/26a/faacm/overview-of-oracle-advanced-controls.html
Oracle Help Center (2026o) How Plan Recommendations Are Released, Release 26A. Supply Planning documentation. Available at: https://docs.oracle.com/en/cloud/saas/supply-chain-and-manufacturing/26a/fausp/how-plan-recommendations-are-released.html
Oracle Help Center (2026p) Overview of Supply Allocation Rules, Release 26B. Backlog Management documentation. Available at: https://docs.oracle.com/en/cloud/saas/supply-chain-and-manufacturing/26b/faubm/overview-of-supply-allocation-rules.html
Oracle Help Center (2026q) Tender Shipment, Release 26B. Oracle Transportation Management documentation. Available at: https://docs.oracle.com/en/cloud/saas/transportation/26b/otmol/planning/shipment_manager/tender_shipment.htm
Oracle Help Center (2026r) Withdraw a Tender, Release 26A. Oracle Transportation Management documentation. Available at: https://docs.oracle.com/en/cloud/saas/transportation/26a/otmol/planning/shipment_manager/withdraw_tender.htm
Oracle Help Center (2026s) Schedule Territory Processes. Fusion Sales documentation. Available at: https://docs.oracle.com/en/cloud/saas/sales/fastr/schedule-territory-processes.html
Oracle Help Center (2026t) Update and Create Territories Using a New Proposal. Fusion Sales documentation. Available at: https://docs.oracle.com/en/cloud/saas/sales/fastr/update-and-create-territories-using-a-new-proposal.html
Oracle Help Center (2026u) When are territories assigned to accounts? Fusion Sales documentation. Available at: https://docs.oracle.com/en/cloud/saas/sales/fasqa/when-are-territories-assigned-to-accounts.html
Oracle Help Center (2026v) How do I assign work to a queue? Fusion Service documentation. Available at: https://docs.oracle.com/en/cloud/saas/fusion-service/fairs/how-you-assign-work-to-a-queue.html
Oracle Help Center (2026w) Why can’t I start a workflow to review a knowledge article? Fusion Service knowledge documentation. Available at: https://docs.oracle.com/en/cloud/saas/fusion-service/fafqa/why-can-t-i-start-a-workflow.html
Oracle Help Center (2026x) Overview of Capacity-Based Quota Management. Oracle Fusion Field Service documentation. Available at: https://docs.oracle.com/en/cloud/saas/field-service/faccu/c-overview-of-capacity-based-quota-management.html
Oracle Help Center (2026y) Migration Package REST Endpoints. REST API Services for Oracle CPQ. Available at: https://docs.oracle.com/en/cloud/saas/configure-price-quote/cxcpq/api-migration-migration-package.html
Oracle Help Center (2026z) Ledger Agentic Application. Financials, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/erp/26d/fins26d/26D-fin-wn-f51915.htm
Oracle Help Center (2026aa) Collector Workspace Agentic Application for enhanced collections actions. Financials, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/erp/26d/fins26d/26D-fin-wn-f51379.htm
Oracle Help Center (2026ab) Workforce Orchestration Command Center. Workforce Modeling & Prediction, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/hcm/26d/womp-26d/26D-workforce-mp-wn-f51902.htm
Oracle Help Center (2026ac) Procurement Spend Command Center. Procurement, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/scm/26d/proc26d/26D-procurement-wn-f51649.htm
Oracle Help Center (2026ad) Simulate award scenarios in the Sourcing Command Center. Procurement, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/scm/26d/proc26d/26D-procurement-wn-f51463.htm
Oracle Help Center (2026ae) Develop Account Territory Growth Plan. Sales Force Automation, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/sales/26d/sfau-26d/26D-sf-automation-wn-f51990.htm
Oracle Help Center (2026af) Strategic Field Demand Planning using Fusion Claw. Field Service, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/service/26d/fscv-26d/26D-fsvc-wn-f52013.htm
Oracle Help Center (2026ag) Use Oracle Fusion Claw to create autonomous outcomes with agentic applications. Release 26D Common Technologies and User Experience What’s New, feature f51374. Available at: https://docs.oracle.com/en/cloud/saas/readiness/common/26d/common26d/26D-common-wn-f51374.htm
Oracle Help Center (2026ah) Use Candidate Sourcing Workspace to Create Candidate Pipelines Before a Requisition Opens. Recruiting, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/hcm/26d/recr-26d/26D-recruiting-wn-f50572.htm
Oracle Help Center (2026ai) Use the Enhanced Candidate Outreach Experience. Recruiting, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/hcm/26d/recr-26d/26D-recruiting-wn-f50574.htm
Oracle Help Center (2026aj) Service Operations Workspace. Service Center, Release 26D What’s New. Available at: https://docs.oracle.com/en/cloud/saas/readiness/service/26d/bsvc26d/26D-b2b-service-wn-f51413.htm
Oracle Help Center (n.d.a) Oracle AI for Fusion Applications: Fusion Claw Capabilities. Documentation page. Available at: https://docs.oracle.com/en/cloud/saas/fusion-ai/fusion-claw.html
Oracle Help Center (n.d.b) Oracle Cloud Applications Readiness, ERP, HCM, SCM, Sales and Service. Readiness pages. Available at: https://docs.oracle.com/en/cloud/saas/readiness/erp.html , https://docs.oracle.com/en/cloud/saas/readiness/hcm.html , https://docs.oracle.com/en/cloud/saas/readiness/scm.html , https://docs.oracle.com/en/cloud/saas/readiness/sales.html and https://docs.oracle.com/en/cloud/saas/readiness/service.html
Oracle blogs and product pages
Oracle Corporation (n.d.a) Oracle Compensation. Oracle Fusion Cloud HCM product page. Available at: https://www.oracle.com/human-capital-management/talent-management/compensation/
Oracle Corporation (n.d.b) Oracle Compensation – Datasheet. Available at: https://www.oracle.com/uk/human-capital-management/talent-management/compensation-datasheet/
Oracle Fusion Development (2025) Quarterly updates made easy. The Fusion Insider blog, December 10. Available at: https://blogs.oracle.com/fusioninsider/quarterly-updates-made-easy
Oracle Fusion Development (2026) Introducing Fusion Claw for complex, autonomous work. The Fusion Insider blog, October 13. Available at: https://blogs.oracle.com/fusioninsider/introducing-fusion-claw-for-complex-autonomous-work
Rajgarhia, G., Satyamurthy, A., Ragul, A., Mathew, R. and Kerr, B. (2026) Identity and Security in AI Agent Studio. Oracle Fusion Development Center of Excellence blog, July 27. Available at: https://blogs.oracle.com/fusioncoe/identity-and-security-in-ai-agent-studio
Sivakumar, R., Tiwari, R., Jimenez, M. and Oracle Application Security (2026) Fusion AI Agents: Secure by Enforcement Best-Practice Considerations. Oracle Cloud Infrastructure blog, July 29. Available at: https://blogs.oracle.com/cloud-infrastructure/fusion-ai-agents-secure-by-enforcement
Oracle investor day
Oracle Corporation (2026j) Investor Day 2026. Transcript, October 28, as published by Stock Analysis. Cited for product and customer statements only. Available at: https://stockanalysis.com/stocks/orcl/transcripts/780561-investor-day-2026/
Named media
du Preez, D. (2026a) Oracle AI World 26 – Fusion is becoming a system of outcomes. Are organizations ready to work that way? Diginomica, October 26. Available at: https://diginomica.com/oracle-ai-world-26-fusion-becoming-system-outcomes-are-organizations-ready-work-way
du Preez, D. (2026b) Oracle AI World 26 – HCM customers want agents, but process, data and legal sign-off come first. Diginomica, October 28. Available at: https://diginomica.com/oracle-ai-world-26-hcm-customers-want-agents-process-data-and-legal-sign-come-first
Essex, D. (2026) Oracle’s agentic AI outpacing customers’ ability to adopt it. TechTarget, October 27. Available at: https://www.techtarget.com/news/oracles-agentic-ai-outpacing-customers-ability-to-adopt-it/872938/
Golden, R. (2026) Revamped Colorado AI law targets ‘consequential’ HR decisions, takes effect in 2027. HR Dive, May 15. Available at: https://www.hrdive.com/news/revamped-colorado-ai-law-targets-consequential-hr-decisions/820425/
Jackson, D. (2026a) Oracle: Fusion agents stop recommending and start doing. ChannelLife Canada, October 26. Available at: https://channellife.ca/story/oracle-fusion-agents-stop-recommending-and-start-doing
Jackson, D. (2026b) Oracle Fusion AI: Autonomy is earned, one outcome at a time. CFOtech, October 29. Available at: https://cfotech.news/story/oracle-fusion-ai-autonomy-is-earned-one-outcome-at-a-time
Lauchlan, S. (2026) Oracle AI World 26 – baking a supply chain global view with AI insight at Grupo Bimbo. Diginomica, October 29. Available at: https://diginomica.com/oracle-ai-world-26-baking-supply-chain-global-view-ai-insight-grupo-bimbo
Reed, J. (2026) Oracle AI World 26 – if Fusion Claw changes enterprise work, does it change cost also? Oracle leaders reveal Fusion Claw economics. Diginomica, October 27. Available at: https://diginomica.com/oracle-ai-world-26-if-fusion-claw-changes-enterprise-work-does-it-change-cost-also-oracle-leaders
Rules, regulators and official texts
Australian Trade and Investment Commission (n.d.) Tax and superannuation. Study Australia. Available at: https://studyaustralia.gov.au/en/work-in-australia/work-rights-and-responsibilities/tax-and-superannuation
California Privacy Protection Agency (n.d.) Frequently Asked Questions on the CCPA regulations, including automated decisionmaking technology. Available at: https://cppa.ca.gov/faq.html
Colorado Attorney General (2026) Artificial Intelligence. Office of the Attorney General page on Senate Bill 26-189, updated October 27. Available at: https://coag.gov/ai/
European Commission (n.d.a) AI Act Service Desk: Annex III, High-risk AI systems referred to in Article 6(2). Consolidated text as at July 27, 2026. Available at: https://ai-act-service-desk.ec.europa.eu/en/ai-act/annex-3
European Commission (n.d.b) AI Act Service Desk: Article 26, Obligations of deployers of high-risk AI systems. Available at: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-26
European Commission (n.d.c) AI Act Service Desk: Article 113, Entry into force and application, as amended. Available at: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-113
European Union (2023) Directive (EU) 2023/970 of the European Parliament and of the Council of 10 May 2023 to strengthen the application of the principle of equal pay for equal work or work of equal value between men and women through pay transparency and enforcement mechanisms. EUR-Lex. Available at: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32023L0970
European Union (2026) Regulation (EU) 2026/1744 (Digital Omnibus on AI), amending Regulation (EU) 2024/1689. Official Journal, July 24. EUR-Lex. Available at: https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202601744
Illinois General Assembly (2024) Bill Status of HB 3773, 103rd General Assembly, Public Act 103-0804. Available at: https://ilga.gov/ftp/legislation/103/BillStatus/HTML/10300HB3773.html
New York City (n.d.) Administrative Code, Section 20-870, Definitions, Automated Employment Decision Tools. American Legal Publishing. Available at: https://codelibrary.amlegal.com/codes/newyorkcity/latest/NYCadmin/0-0-0-135839
New York City Department of Consumer and Worker Protection (n.d.a) Rules of the City of New York, Title 6, Section 5-300. American Legal Publishing. Available at: https://codelibrary.amlegal.com/codes/newyorkcity/latest/NYCrules/0-0-0-138393
New York City Department of Consumer and Worker Protection (n.d.b) Automated Employment Decision Tools. Agency page. Available at: https://www.nyc.gov/site/dca/about/automated-employment-decision-tools.page
Office of the New York State Comptroller (2025) DiNapoli: New Yorkers Deserve a Transparent Hiring Process When Artificial Intelligence Is Used To Vet Their Job Applications. Press release, December 2. Available at: https://www.osc.ny.gov/press/releases/2025/12/dinapoli-new-yorkers-deserve-transparent-hiring-process-when-artificial-intelligence-used-vet-their
Research this note builds on
Greyhound Consilium (2026) The Delegation Dial – How Much Authority To Hand Fusion Claw, Process By Process. Practical guide, developed in partnership with Oracle, October 23. Available at: https://greyhoundresearch.com/the-delegation-dial-how-much-authority-to-hand-fusion-claw-process-by-process/
Greyhound Research (2026a) Everyone Is Building AI Agents. Oracle Is Moving The Contest Inside The Runtime. Research report, July 20, including findings from Greyhound CIO Pulse 2026. Available at: https://greyhoundresearch.com/everyone-is-building-ai-agents-oracle-is-moving-the-contest-inside-the-runtime/
Greyhound Research (2026b) Oracle Fusion Claw – Autonomy Arrives With A Receipt. Research report, October 14. Available at: https://greyhoundresearch.com/oracle-fusion-claw-autonomy-arrives-with-a-receipt/
Greyhound Research (2026c) Oracle Agentic Implementation – When Agents Draft The Setup, Review Becomes The Control. Research note, October 29. Available at: https://greyhoundresearch.com/oracle-agentic-implementation-when-agents-draft-the-setup-review-becomes-the-control/
Greyhound Research (2026d) 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/
Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright
The evidence in this research note is current through October 30, 2026. Any written response received from an organization named in it after that date 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. Greyhound Research did not offer any named organization a pre-publication factual review of this research note. Editorial control remained entirely with Greyhound Research throughout. 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. Oracle is a client of Greyhound Research (part of The House Of Greyhound). The engagement covers only thought-leadership content published under the Greyhound Consilium brand and did not fund this research note. Greyhound Research holds no other financial interest in Oracle or its subsidiaries.
Intended Use
This research note is written for chief information officers, chief financial officers and controllers, chief human resources officers and the heads of payroll and talent acquisition, chief supply chain officers and their planning, procurement, and logistics leads, chief revenue officers and heads of service, general counsel and employment counsel, internal audit leaders, and the press. It reads Oracle’s Fusion Claw announcements at Oracle AI World 2026 function by function, down to the Fusion modules beneath each application, to inform ownership, sequencing, signature, legal mapping, budgeting, and governance decisions. 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 how rules may apply to each application is to be confirmed with counsel for each jurisdiction. Its assessments rest on the documents listed in Sources And References as they stood on October 30, 2026, and can change as Oracle, legislators, regulators, and others publish more. Product availability, features, and terms are described as Oracle published them through October 30, 2026, so confirm them with your Oracle account team before acting on them.
Research Methodology
Greyhound Research’s analysis rests on source evidence verified against the document that supports each claim. The evidence comprises the eighty-four sources listed in Sources And References, in seven classes, from Oracle’s announcements, documentation, blogs, and investor day to named media, the rules and official texts cited, and the research this note builds on. A finding that something is not described or not found is bounded to the documents the text names. Oracle’s Release 26D readiness guides were checked on October 30, 2026 for a feature page that describes each application as Claw-powered, and a page not found is bounded to those guides on that date. The count of application names states the names as Oracle wrote them, compares Oracle’s releases and its Help Center page, and infers no renames.
Where the note cites Oracle’s documentation for a Fusion module, it describes the module as Oracle documents it, and it does not assume how a Claw application uses that module beyond what Oracle states. Release 26D production dates are derived from Oracle’s published cohort rule and its update naming, as in Greyhound Research’s research report of October. The mapping of rules to applications in Table 3 is Greyhound Research’s reading of the official texts cited, to be confirmed with counsel, and is not legal advice.
Every executive and customer quoted is named and linked where the quotation appears. Customers are quoted in their own words as reported from Oracle AI World by named media, and no composite or invented customer appears. Statements from Oracle’s investor day are cited for product and customer facts only, and no financial figure from it is used. The Greyhound CIO Pulse 2026 figure is cited as published in Greyhound Research’s research report of July 20, 2026. The owners, co-signers, chains, readings, and proposals in Tables 3 and 4 and Exhibits 2 to 12 are Greyhound Research’s analysis, not statements by Oracle. No Greyhound Fieldnotes are cited, and no factual claim in this research note rests on unpublished material. Quotations are checked against the live source before publication.
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.