The Delegation Dial – How Much Authority To Hand Fusion Claw, Process By Process

The Delegation Dial: Authority Granted In Advance, Outcome By Outcome

Every enterprise has work too complex for a simple rule and too frequent to do well by hand: closing the books, staffing a ward, consolidating inbound loads, redrawing sales territories. Teams take it to spreadsheets for hours, and the answer is often stale by the time it is ready.

Oracle Fusion Claw, introduced on September 29, 2026, is built for exactly this work. It lets Fusion’s agentic applications take an objective, plan how to reach it, run the calculations at scale, and carry out the result inside Fusion, within the authority the business has delegated. The question is no longer whether AI can do the work, but how much of it to hand over, and for which outcomes. Oracle CEO Mike Sicilia put the division of work in one sentence: “By letting AI take on more of the work, while people set the objectives and guardrails, organizations can unlock tremendous capacity for their teams to focus more on driving growth, innovation, and serving customers.”

Greyhound Consilium developed this guide in partnership with Oracle. It builds on Oracle Fusion Claw – Autonomy Arrives With A Receipt, the independent research report Greyhound Research published on October 14, 2026, and turns it into next steps for the CIO and the leaders who own each process in finance, HR, supply chain and sales. Most parts show the idea in Oracle’s published examples, give a tool to use the same week, and close with Greyhound Consilium’s advice.

Where Fusion Claw sits

Oracle’s technical brief describes a move from systems of record to systems of outcomes, where software completes the work and people step in by exception. Fusion delivers AI in four ways, each more autonomous than the last: generative AI in existing screens, answer agents, workflow agents, and agentic applications that own an outcome end to end. Fusion Claw, which Oracle’s release defines as “a governed agentic execution runtime for Oracle Fusion Agentic Applications that combines AI reasoning with deterministic enterprise computation to enable highly complex work to be economically completed at scale,” lets the fourth kind take on the hardest work. A frontier model plans, a separate layer carries out the plan the same way every time, and the system learns from each cycle. In the brief’s phrase, it “reasons once, then executes continually.” Exhibit 1, carried forward from the research report, puts Oracle’s vocabulary in plain words.

Three of the eight terms work together across the life of every outcome. The Enterprise Operating Envelope is set before anything runs: what the outcome may do, within which limits, and who must agree. The Outcome Trust Harness governs each run as it happens, through identity, capabilities, data and actions. The Outcome Receipt records what happened afterwards. Read in that order, they answer the three questions a board, a regulator or an auditor will ask of any autonomous outcome: who allowed it, what kept it in bounds, and what it did.

Authority, granted in advance

The system published the roster because it fell within the authority the charge nurse had granted in advance, so control stayed with her throughout. In the brief’s words: “Reasoning produced options; authority, granted deliberately and in advance, decided which could run.” Fusion Claw is not a switch. It is a dial that sets how much authority to hand over, for which work, turned up as confidence grows.

Three settings on the dial

Oracle’s brief is clear that “Customers decide how much autonomy to delegate.” Exhibit 2 sets out its three settings in Oracle’s words, with Greyhound Consilium’s guidance on where each fits.

The guardrails never move with the dial. Oracle’s Release 26D documentation says it plainly: “Although AI can plan and recommend actions, it can’t grant itself authority to act.” Oracle’s technical brief adds that moving toward autonomy changes how often a person steps in, “not whether controls apply.”

What does change is the reviewer’s job. At human in the loop, a person approves each proposed action, so the speed of review sets the pace. At human in the lead, the system runs and the person directs it, stepping in where judgment is needed. At autonomous execution, attention moves to exceptions and to the receipts. Oracle’s two worked examples sit at the upper settings: its brief labels the hospital roster human in the lead, and the inbound load plan governed and fully autonomous. Both still routed some items to a person: five loads held for review and six orders sent to a planner in one, and attendance patterns brought to the charge nurse in the other. That is why the dial is best set outcome by outcome.

Decide outcome by outcome

The dial need not sit in one place for a whole application, as Oracle’s own governance guidance shows.

Fusion Claw brings the same thinking to larger outcomes. In AI Agent Studio, autonomous outcomes are enabled for an application and its autonomy settings configured. The fastest movers will decide one level finer: which class of outcome may run at which setting, and who signs for it.

↑ Back to the section guide

What To Have Ready: Six Preparations Before Any Outcome Moves Up The Dial

Most of what a first run needs already exists in Fusion. Greyhound Consilium groups the preparation under six headings.

Authority. The Enterprise Operating Envelope is set in controls teams already know: agent roles, granted permissions, approvals on the tools that act, and policy models that turn written policy into tested logic before it is published. Part 4 shows how to set them.

Release. Oracle introduced 25 Claw-powered applications on September 29, and Fusion Claw ships with Release 26D, which reaches customers in phases, cohort by cohort. Under the cohort rule Oracle published for quarterly updates, production environments take an update on the third Friday of the update month, which for 26D falls on November 20 or December 18, 2026, or January 15, 2027, depending on cohort. Test environments take it two weeks earlier, and Middle East regions a day earlier. Map your dates, any early access, and the applications in your cohort with your Oracle account team.

Budget. Planning draws on AI Units. Oracle’s contract includes 30,000 a month per Fusion production environment, and an Agentic Applications subscription adds a pooled allocation, all counted across development, test and production. Release 26D’s budget controls set budgets per application or agent, send warnings, “Enable auto-stop to prevent new runs when a selected budget is exhausted,” and log threshold events, budget changes and control actions. Oracle’s playbook adds the right habit: set a consumption-cost target for each use case before launch and compare actual use against it. The research report explains how the meter works.

Data and region. Claw-powered applications plan with frontier models from providers such as OpenAI and Google, or with a model the customer brings. Oracle’s FAQ sets out where OpenAI models process requests for each Fusion region. For regulated data, choose the model for each outcome with your Oracle account team, record where its reasoning runs for your Fusion region, and carry both into your contract terms. Check the data too: in Oracle’s shipment example, six orders went to a planner because of setup or master-data issues.

Assurance. Each outcome run produces an auditable, traceable receipt: what was done, on what evidence, under which authority. Brief the audit committee chair and the head of internal audit before the first run.

People. In the EU, the AI Act requires deployers to inform workers’ representatives and the workers affected before certain workplace AI systems are used, from December 2, 2027. In the Netherlands, union guidance for works councils treats AI that plans rosters as a matter for the works council’s consent. Where that applies, the conversation comes before the dial moves. Part 9 covers the rules across seven jurisdictions.

Tool 1 turns the six headings into fourteen items, each with an owner, and sorts them by two gates. Ten items protect the test itself and must be in place before the first test run: among them an owner, a first delegation record entry, a mapped release date, a budget with auto-stop on, checked data, the audit committee chair briefed, and the people affected informed, with consent where it is required. Four protect the move up and must be ready before any outcome moves past human in the loop: the governing policy tested and published, approvals before the actions that commit, receipt review agreed, and the correction routine rehearsed. The first gate is mostly decisions, so it moves as fast as the people involved. The second draws on what the test runs show, so it closes later.

↑ Back to the section guide

The First Process: Autonomy For The Plan, A Person For The Commitment

The first outcome sets the pace: when the people who will approve the next one see it succeed, every later step is easier.

Oracle’s advice is to look at recurring work that takes significant effort to investigate, plan and carry through, then decide where Fusion Claw can take on more of it while people stay in control. Oracle’s release calls it “specialist-grade work including deep research, computation, simulation, modeling, and continuous re-planning.” Oracle’s playbook adds a filter: rank candidates by return target, volume and the number of people who benefit, and baseline today’s work.

Separate the plan from the commitment

Some actions can be reworked freely. Others, such as a cleared payment or a departed shipment, are final. The strongest first candidates let planning run with real autonomy while the final commitment stays with a person. Correction happens through Fusion’s own transactions, so design the commitment step around how your team corrects results today.

That makes a supply chain plan a natural first outcome for many organizations, in line with the research report’s advice: operations first, with finance and HR following once the first receipts have been read.

Tool 2 scores candidates on six tests: volume, complexity, correction, data, baseline and owner. Two rules come first. An outcome whose actions cannot be undone does not go first, and neither does an outcome with no baseline, because no one can show the gain.

Oracle’s inbound load plan shows what a strong candidate looks like on the scorecard. The work is complex, with hundreds of orders, many constraints and a pull-forward limit. The baseline is measured, at 61 percent utilization against a 70 percent target. The plan can be reworked before anything is released, and the few orders with data issues go to a planner rather than stopping the run. A candidate that scores well on correction and baseline but poorly on data is often worth a cycle of cleanup first, because data issues send work to people instead of to the plan.

↑ Back to the section guide

The Enterprise Operating Envelope: Authority Set In Controls Fusion Already Holds

Oracle describes the Enterprise Operating Envelope as “the authority layer that defines the boundaries for the work.” Release 26D documentation says Fusion Claw “operates within the organization’s established policies, permissions, approvals, risk thresholds, and authority limits,” so the envelope is set in controls Fusion already holds, and Fusion Claw works inside it by design.

The research report puts it plainly: “The envelope is the delegation matrix made executable. The receipt is the evidence that it was obeyed.” The schedule every finance function keeps, of who may approve what, up to what value, under which policy, becomes something software must obey.

How the work is separated

Oracle built Fusion Claw so that the part that thinks is never the part that decides what is allowed. Exhibit 3 shows the four layers in its brief. Each outcome also runs “inside an isolated, temporary execution environment” that is “created for that outcome and torn down when it completes,” so it “cannot reach beyond its assigned scope, and leaves no standing process behind.” Oracle’s brief adds that permitted plans “are carried out deterministically within Fusion Applications,” and that “responsibility for governed execution stays in the Fusion layer.”

The separation matters because it fixes where trust has to be earned. A better model can make better plans, but it cannot widen what is permitted, because authority sits in a different layer that the business sets. The execution layer works the same way every time, so a permitted plan runs as permitted. Enforcement sits apart from all three, so a higher setting on the dial never switches the controls off. For a CIO, that puts the review effort where it belongs: in the authority the business grants and in the receipts that show how it was used.

The controls that make up the envelope

Six controls hold the envelope in Fusion, four that set the authority and two that keep each run in hand. Who may act: agents hold role-based access, and Fusion remains the authorization authority, so using an agent never widens anyone’s access. What the rules say: policy models turn written policy into decision logic that teams test before they publish it. When a person must agree: approval steps sit before the actions that post or commit, and Fusion Claw “requests approval when required, and escalates exceptions when necessary.” How far it may go: the application’s autonomy settings. The other two keep each run in hand: what it may spend, through the AI Units budgets and auto-stop, and what happens on an exception, through its escalation path. Each is set out in an Oracle document a team can open today, and Exhibit 4 carries the research report’s list forward as a reading list.

The same principle already runs in Fusion’s workflow agents, the third of the four ways Fusion delivers AI, and many teams will recognize it from there.

The invoice agent shows the envelope at small scale. Policy defines the options, the agent chooses among them, and one of those options is to prepare the case for a person to decide. Fusion Claw applies the same pattern to outcomes that span many orders, shifts or accounts at once, so a team that already governs its workflow agents well has most of the habits Fusion Claw needs.

The receipt, and the record that goes with it

Every run produces a receipt. In Oracle’s shipment example, it records the requester, owner, executor and approver, the governing policy, data sources, linked transactions, a verification ID and a digital signature. Pair it with a short delegation record, one entry per outcome: what may run, at which setting, up to what threshold, under which policy version, delegated by whom, and whose second signature widens it. The receipt records the authority applied to each run, and the record names who granted it and when, so a reviewer can check one against the other.

Oracle’s Fusion Center of Excellence advises that “a material change to rules, prompts, schemas, or templates deserves a new test run before publication.” Tool 3 maps each part of the authority to Fusion, then gives the record as a template.

Keep the record short. One entry per outcome is enough, and the worked example shows the level of detail that works: a sentence for the outcome, a line for the limit, the policy and its version, and the names. Write the first entry before the first test run, read it beside the first receipts, and resolve any difference between them, with the second signature for any change that widens what runs. By the time the outcome is ready to move up, the record and the receipts should tell the same story, and that agreement is what the second signature confirms.

↑ Back to the section guide

Process By Process: Where The Dial Starts In Each Function

Of the 25 Claw-powered applications in Release 26D, Oracle leads with four, one per function. Its release calls them Ledger, Workforce Staffing, Shipping Consolidation and Account Territory Growth Plan, and its blog of October 13 describes the same functions as Ledger Workspace, Workforce Operations Command Center, Shipment Load Planning Workspace and Account Territory Growth Plan. Your Oracle account team can confirm which applies in your cohort. Oracle’s Help Center describes fifteen capabilities so far, as of mid-October 2026, and Exhibit 5 reads each as a use case.

Tool 4 gives a process card for each of the four, and every card answers the same eight questions, so the governance council can read them side by side. The first outcome on each card is chosen so a reviewer can judge it in one cycle. The line on what stays with a person at first names the commitments the owner keeps for now, and any widening later needs the second signature. The signal to move the dial names the evidence in advance, so the decision to move up becomes a check against an agreed bar rather than a fresh debate.

Finance: closing the books with Ledger

Ledger reconciles large volumes of entries and investigates anomalies. The Help Center adds that it analyzes account variances, checks account usage against controls and examines account activity before the close, guided by the customer’s own standard operating procedures. Related applications widen the view, as Oracle’s own description of one shows.

HR: staffing with Workforce Staffing

Workforce Staffing builds executable staffing plans that balance skills, accreditations, availability, labor rules and costs, and re-plans as conditions change. The Help Center’s schedule balance simulations compare scenarios when a schedule misses fill-rate, budget or policy goals, while preserving hard constraints. Oracle’s hospital example in part 1 shows the result.

Workforce outcomes also carry the people and rules work in parts 8 and 9.

Supply chain: fuller loads with Shipping Consolidation

Shipping Consolidation helps reduce shipping costs while addressing service commitments, modeling consolidation options and, when authorized, executing the best-supported plan. The Shipment Load Planning Workspace generates rough-cut load plans from equipment profiles, forecast volumes and constraints, and Oracle’s example in part 3 is the model. Nearby, the Procurement Spend Command Center finds opportunities to redirect purchases to existing agreements, and the Sourcing Command Center simulates awards against your own sourcing policies.

Sales: coverage with Account Territory Growth Plan

Account Territory Growth Plan compares territory scenarios and weighs tradeoffs against capacity. The Help Center adds that it combines customer, pipeline and market data into a prioritized growth plan for every account. Oracle’s governance guidance gives the pattern: sales agents carry approval checkpoints for customer-facing changes.

↑ Back to the section guide

The First 90 Days: From A Named Outcome To A Higher Setting

A first quarter should end with one outcome at a higher setting than it started, a measured gain, and a routine the next outcome can reuse. The order follows the research report’s runbook: set the authority and budget, agree the model and record where its reasoning runs, sort actions by how they are corrected, read the receipts, and measure units per outcome before setting the budget. The test runs need Release 26D in a non-production environment, which the cohort rule brings two weeks before your production date, or early access, so plan the weeks with your Oracle account team. Oracle has said it expects organizations to begin with more human review and expand automation as they gain experience, and its own governance guidance follows the same rhythm.

Oracle’s playbook adds how to widen the circle: expand in waves of roughly ten, fifty and two hundred users, check quality, spend and returns between waves, and where possible keep the existing process running until confidence builds. The research report reaches the same first step: begin where the plan is staged for review, and move one outcome class at a time toward autonomy once the person who answers for it has read its receipts.

Tool 5 follows Oracle’s pattern closely. Weeks 1 to 4 match Oracle’s first 30 days: charter the council, agree the thresholds, and pick the outcome. Weeks 5 to 10 cover the end-to-end runs and the autonomy decision that Oracle places between day 30 and day 60, with the approval points already set in weeks 3 and 4. Weeks 11 to 13 match the last stretch: review the early outcomes and set the rhythm for the next ones. The reading of receipts in weeks 5 to 8 is where most of the learning happens, so protect the reviewers’ time in those weeks.

↑ Back to the section guide

What To Measure: The Gain, The Controls, And The Units Behind Them

What gets measured gets delegated. A business that can show the first outcome’s gain, in numbers its CFO already watches, will find the second easy to approve.

Oracle’s playbook sets the standard: “The goal is a portfolio you can defend with numbers.” Track outcomes completed, exceptions avoided and shorter cycle times, anchor value in avoided cost, and review results and costs regularly in AI Agent Studio. Put two kinds of measure on one page: business measures that show the gain, and control measures, such as items routed, held or overridden, receipts read and units per outcome, that show it was earned within the authority granted. The control measures are what move an outcome up the dial. Tool 6 sets Oracle’s example figures beside columns for your own.

Read the scorecard in pairs. In Oracle’s hospital example, coverage rising from 70 to 99 percent is the business measure, and a receipt with no policy exceptions is the control measure that shows the gain was earned within the charge nurse’s authority. In the load plan, utilization rising to 88.2 percent is the gain, and the six orders routed and five loads held show the controls at work. A gain without its control measure is a claim, and a control measure without a gain is a cost. Together they are what a CFO and an auditor will both sign for.

Budgeting AI Units

Oracle’s brief describes paying for trusted execution “described in outcome terms rather than component or token math.” Oracle’s contract measures usage in AI Units, and Oracle’s Fusion Insider blog states that “an AI Unit is worth one cent.” Plan in AI Units, and work with your Oracle account team on how outcome-based terms apply to your agreement.

Each metered action counts the tokens a model reads and writes, divided by 10,000 and rounded up, times a factor: for a Fusion Claw action, 20 on a Frontier model and 10 on a model you bring. Oracle’s release explains why planning is where the cost sits: separating reasoning from execution “provides Fusion Applications customers with a potential economic advantage by applying AI reasoning only where intelligence is needed and using efficient deterministic computation for high-volume execution.” Oracle’s Fusion Insider blog explains that its two-part approach “can lower costs because AI is used only for the parts of the work that require reasoning,” and Oracle’s technical brief adds that because Fusion Claw “reasons once, then executes continually,” customers “don’t pay the reasoning cost on every action.” Planning is where most of the cost sits. Simple agent steps around the plan, such as approving transactions, count at zero on the Basic model hosted on Oracle Cloud Infrastructure and at 20x on a Frontier model, under Oracle’s contract. So measure units per outcome from your own runs. Tool 7 sets out the arithmetic, and the research report sets out the meter’s full arithmetic. The contract allows its factor tables to change on at least thirty days’ notice, so revisit the budget each quarter and whenever Oracle gives notice of a change.

Three habits keep the budget honest. Measure units per outcome in the test runs rather than estimating them, because the same outcome can draw different units depending on how much planning it needs. Count everything else that draws on the same allocation, since the included units are shared with every other AI use in the environment. And set the warning threshold well below the cap, so the owner hears about a change in consumption before auto-stop has to act.

↑ Back to the section guide

Who Answers For The Outcome: Roles, Records, And The People Around Them

Fusion Claw changes who does the work, not who answers for it. Oracle’s governance guidance is direct: “the people best positioned to govern AI capabilities are those who own the outcomes.”

Tool 8 puts names against each decision.

Two habits make the map work. No row is complete until it carries two names, because a decision with one name has no second check. And the people who set a limit should not be the only ones to test it: the process owner and the council set the authority, and internal audit and the controller also check how it was used. Oracle’s three layers give the same division at a higher level, with IT owning the controls every agent inherits, the business owning its domain, and the council deciding where the line falls for each use case.

Three records that make it auditable

The first is the delegation record in Tool 3. The second, Tool 9, sorts every action by how it is corrected, which sets where it goes first once its receipts have been read. Oracle’s brief is clear that “Customers decide how much autonomy to delegate,” so the owner widens a setting through the delegation record, with its second signature, as confidence grows. An outcome that changes people’s work or pay can start lower, as the process cards and Part 9 advise. Correction is handled within Fusion, through its own transactions, so the register makes every outcome’s corrections, and who makes them, known before it runs.

The third is the receipt, read on a routine. Oracle has said the receipt carries more context than a conventional audit log, including the full text of the policies used during the task. Tool 10 sets out the checks for the people who read it.

During the test runs, read every receipt. Once an outcome runs at a higher setting, a sample is enough, sized by internal audit and weighted toward runs with exceptions or items held. The checks marked for the controller and internal audit are the ones an external auditor is most likely to repeat, so agree their evidence early.

The people side

In Oracle’s examples, people stay at the center of the work, and what changes is the work they do. The charge nurse spent her time on the one decision that needed her. A planner took the six orders with data issues, while the system built 154 loads and held five for review. That is the shape of the work ahead for planners, schedulers, accountants and sales operations analysts: setting limits, reading receipts, and handling the exceptions that need judgment.

↑ Back to the section guide

The Rules In Practice: What Regulators Ask And What Fusion Claw Provides

The research report reads the rules in seven jurisdictions and finds them converging on a four-part rule: the organization answers for the agent, a competent person can halt it before its output takes effect, the record lets a third party reconstruct what it did, and an affected person can reach someone empowered to change the result. Fusion Claw was built with the same instincts. Exhibit 6 sets each part beside what Fusion Claw provides and what the organization adds.

Workforce outcomes come first everywhere, because Annex III of the AI Act reaches systems used “to allocate tasks based on individual behaviour or personal traits or characteristics.”

Auditors are moving too. The proposed revision of the international standard on audit evidence, open for comment until December 15, 2026, adds authenticity to the attributes of the reliability of information used as audit evidence. Tool 11 gives each jurisdiction’s rule, its date, and what to do.

Use the runway backwards. Take each date, list the outcomes it touches, and set the work that must be finished before it. The nearest date still to come is Australia’s transparency duty on December 10, 2026, which calls for a privacy policy that describes the decisions a computer program makes, or substantially and directly supports, that could significantly affect an individual’s rights or interests, including any that Claw outcomes make or support. US state rules follow from January 1, 2027, India’s data protection rules reach full force in May 2027, eighteen months after notification, and the EU’s duties for Annex III workforce systems apply from December 2, 2027. Most of the work is the same everywhere: a named owner, a way to halt, a record that reconstructs each run, and a route for an affected person to someone who can change the result.

↑ Back to the section guide

From One Outcome To Many: Autonomy That Expands As Receipts Prove It

The first outcome proves the routine, so the next ones move faster. Oracle’s governance guidance says it plainly, of quotes and pricing: “Autonomy expands only when performance and controls are proven.”

Fusion Claw also improves with use. Oracle’s brief explains that it re-plans within a job’s authority and, across runs, “carries forward what worked.” As enterprises bring their own agents and assistants, governed execution stays in Fusion whichever client made the request, with Fusion Claw supplying approval chains, delegation rules, separation of duties and transaction integrity. Oracle’s playbook closes the loop: reinvest demonstrated savings in the next wave, because “A disciplined program is one that knows when to double down and when to step back.” Tool 12 is the quarterly review that makes those calls.

The review works best as a standing meeting with a fixed agenda. Each question names the evidence to bring, so the council decides from the scorecard, the receipts and the record rather than from impressions. Over time, the review also builds the organization’s own library of patterns: each vetted outcome becomes a template the next one starts from, as Oracle’s governance guidance advises, so similar outcomes can follow its expedited review.

↑ Back to the section guide

Getting Started With Oracle: Four Working Sessions To A Dated Plan

Everything in this guide is designed to be done with your Oracle account team and implementation partner. They know your cohort, applications and configuration. Your teams know your outcomes, policies and people. Four working sessions bring the two together and fit within the first month of the 90-day plan.

Each session leaves something the next one needs, so book all four at once. The map gives the second session its outcomes, the second gives the third its controls, and the third gives the fourth its budget and model. Bring the people who will own the decisions, not only those who will configure them: the process owner to every session, the CISO, the data protection officer and finance to the third, and the controller and internal audit to the fourth.

↑ Back to the section guide

Terms Used In This Guide

Agentic application. Software that owns a whole outcome from start to finish, rather than one step of it.

AI Agent Studio. Where administrators enable Claw-powered applications and set their autonomy settings, policy models, approvals and budgets.

AI Units. Oracle’s measure of AI usage: for each metered AI action, the tokens a model reads and writes, divided by 10,000 and rounded up, multiplied by a factor for the model tier and the kind of action. Deterministic execution under an approved plan draws none.

Autonomous execution. The setting on the dial where the system runs within its guardrails and people handle exceptions.

Cohort. The group a production environment belongs to, which sets which of Oracle’s three dates it takes each quarterly update on.

Delegation record. Greyhound Consilium’s term for a short record, one entry per outcome, of what was delegated, at which setting, under which policy version, and by whom.

Deterministic computation. Rule-following calculation, where the same inputs give the same results every time.

Enterprise Operating Envelope. Oracle’s authority layer: the objectives, policies, procedures, permissions, approvals and escalation points a business sets, held in controls Fusion already has.

Frontier model. The most capable class of AI model, from providers such as OpenAI and Google, which Fusion Claw can use to plan, alongside a model the customer brings.

Human in the lead. The setting on the dial where the system runs and a person directs it and steps in.

Human in the loop. The setting on the dial where a person approves each action before it runs.

Outcome Receipt. The auditable record of a run: the authority applied, the evidence used, the decisions made, the actions and transactions, and the result.

Outcome Trust Harness. The controls Oracle applies to each run, covering identity, capabilities, data and actions.

Policy model. A written policy turned into tested decision logic, published in AI Agent Studio before it is used.

Release 26D. The Fusion quarterly update that brings Fusion Claw.

Runtime. The engine that actually runs the work.

Workflow agent. An agent that carries out multi-step tasks within rules and approvals, adding reasoning where the rules run out.

↑ Back to the section guide

Sources And References

Every fact, figure and quotation in this guide traces to one of the sources below, presented in Harvard reference format and grouped by class. In the web edition, each source is also linked in the sentence that carries it. All sources were accessed between October 5 and October 15, 2026.

Oracle announcements and briefs

Oracle Corporation (2026a) 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 (2026b) Oracle Fusion Claw. Technical brief, Version 1.0, October. Available at: https://www.oracle.com/a/ocom/docs/oracle-fusion-claw-technical-brief.pdf

Oracle Corporation (2026c) Oracle Fusion Cloud Service Descriptions. Contract document, v100826, effective October 8. Available at: https://www.oracle.com/contracts/docs/oracle-fusion-cloud-service-desc-1843611.pdf

Oracle Corporation (n.d.) Oracle AI for Fusion Applications. Product page. Available at: https://www.oracle.com/applications/fusion-ai/

Oracle documentation

Oracle Help Center (2026a) Manage AI unit budget, consumption, and controls. Release 26D Common Technologies and User Experience What’s New, feature f51348, updated September 21. Available at: https://docs.oracle.com/en/cloud/saas/readiness/common/26d/common26d/26D-common-wn-f51348.htm

Oracle Help Center (2026b) Where are the locations of the OpenAI endpoints for AI agents in Fusion Applications? Oracle AI for Fusion Applications Questions and Answers, build stamp September 22. Available at: https://docs.oracle.com/en/cloud/saas/fusion-ai/aiaqa/where-are-the-locations-of-the-openai-endpoints-for-ai-agents-in-fusion-applications.html

Oracle Help Center (2026c) Add a Policy Model to a Policy Node of Workflows. Release 26D guide, September 24. Available at: https://docs.oracle.com/en/cloud/saas/fusion-ai/26d/aiaas/add-a-policy-model-to-a-policy-node-of-workflows.html

Oracle Help Center (2026d) Use Oracle Fusion Claw to create autonomous outcomes with agentic applications. Release 26D Common Technologies and User Experience What’s New, feature f51374, delivered September 28. Available at: https://docs.oracle.com/en/cloud/saas/readiness/common/26d/common26d/26D-common-wn-f51374.htm

Oracle Help Center (2026e) Oracle AI for Fusion Applications: Fusion Claw Capabilities. Documentation page, as read on October 15. Available at: https://docs.oracle.com/en/cloud/saas/fusion-ai/fusion-claw.html

Oracle Help Center (2026f) How do I configure the retention period for generative AI data? Oracle AI for Fusion Applications Questions and Answers, build stamp September 22. Available at: https://docs.oracle.com/en/cloud/saas/fusion-ai/aiaqa/how-do-i-configure-the-retention-period-for-generative-ai-data.html

Oracle blogs and demonstrations

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 (2026a) New Fusion Agentic Applications—details and demos. The Fusion Insider blog, May 15. Available at: https://blogs.oracle.com/fusioninsider/new-fusion-agentic-applications-details-and-demos

Oracle Fusion Development (2026b) A playbook for scaling business value with Fusion AI. The Fusion Insider blog, August 27. Available at: https://blogs.oracle.com/fusioninsider/a-playbook-for-scaling-business-value-with-fusion-ai

Oracle Fusion Development (2026c) Workflow agents—where business rules meet AI reasoning. The Fusion Insider blog, September 17. Available at: https://blogs.oracle.com/fusioninsider/workflow-agents-where-business-rules-meet-ai-reasoning

Oracle Fusion Development (2026d) 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

Oracle Fusion Development (2026e) Who owns AI governance? The Fusion Insider blog, October 13. Available at: https://blogs.oracle.com/fusioninsider/who-owns-ai-governance

Rajgarhia, G. (2026) Policy Model in Workflow Agents. Oracle Fusion Development Center of Excellence blog, August 5. Available at: https://blogs.oracle.com/fusioncoe/policy-model-in-workflow-agents

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

Oracle executives in named media

Gillin, P. (2026) Oracle expands AI agent features with Fusion Claw. SiliconANGLE, September 29. Available at: https://siliconangle.com/2026/09/29/oracle-expands-ai-agent-functionality-with-fusion-claw/

Research this guide builds on

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/

Regulator, government and standard-setter texts

Board of Governors of the Federal Reserve System, Office of the Comptroller of the Currency and Federal Deposit Insurance Corporation (2026) SR 26-2, Revised Guidance on Model Risk Management. April 17. Available at: https://www.federalreserve.gov/supervisionreg/srletters/SR2602.htm

California Privacy Protection Agency (2025) Regulations on cybersecurity audits, risk assessments and automated decisionmaking technology, approved text, with automated decisionmaking requirements applying from January 1, 2027. Available at: https://cppa.ca.gov/regulations/pdf/ccpa_updates_cyber_risk_admt_appr_text.pdf

Colorado Attorney General (2026) Artificial intelligence rulemaking page, law effective January 1, 2027. Available at: https://coag.gov/ai/

European Union (2024) Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 26 and Annex III, as amended by Regulation (EU) 2026/1744, applying to Annex III systems from December 2, 2027. Text as published by the AI Act Explorer. Available at: https://artificialintelligenceact.eu/article/26/ and https://artificialintelligenceact.eu/annex/3/

Infocomm Media Development Authority (2026) Model AI Governance Framework for Agentic AI, Version 1.0. January 22. Available at: https://www.imda.gov.sg/-/media/imda/files/about/emerging-tech-and-research/artificial-intelligence/mgf-for-agentic-ai.pdf

International Auditing and Assurance Standards Board (2026) Proposed International Standard on Auditing 500 (Revised), Audit Evidence, exposure draft. August, comments due December 15. Available at: https://ifacweb.blob.core.windows.net/publicfiles/2026-08/IAASB-Audit-Evidence-Risk-Response-ISA-500-Exposure-Draft.pdf

Murmu, S. C. (2026) Trusted Innovation: Shaping the Future of Finance. Speech, Reserve Bank of India, Global FinTech Fest, Mumbai, September 11. Available at: https://www.rbi.org.in/Scripts/BS_SpeechesView.aspx?Id=1579

Office of the Australian Information Commissioner (2026) New resources on transparency for use of AI and automated decision-making. September 30. Available at: https://www.oaic.gov.au/news/media-centre/new-resources-on-transparency-for-use-of-ai-and-automated-decision-making

Office of the Privacy Commissioner, New Zealand (n.d.) Principle 8, Privacy Act 2020. Available at: https://www.privacy.org.nz/privacy-act-2020/privacy-principles/8/

Outlook Money (2026) Sebi to Issue AI/ML Guidelines for Capital Markets: Mandatory ‘Kill Switch’ and Human Oversight Framework Explained. Named media report of a regulator’s statement of August 19, updated August 20. Available at: https://www.outlookmoney.com/invest/sebi-to-issue-aiml-guidelines-for-capital-markets-mandatory-kill-switch-and-human-oversight-framework-explained

Press Information Bureau, Government of India (2025) Government notifies DPDP Rules to empower citizens and protect privacy. November 14. Available at: https://www.pib.gov.in/PressReleasePage.aspx?PRID=2190014&reg=3&lang=2

Reserve Bank of India (2026) RBI issues draft Guidance on Regulatory Principles for Model Risk Management. Press release, June 24. Available at: https://www.rbi.org.in/scripts/BS_PressReleaseDisplay.aspx?prid=63006

United Kingdom (2025) Data (Use and Access) Act 2025, section 80, automated decision-making provisions in force February 5, 2026. Available at: https://www.legislation.gov.uk/ukpga/2025/18/section/80

Employee representatives

FNV, ARBO netwerk (2025) Medezeggenschap en AI. Guidance for works councils, December. Available at: https://www.fnv.nl/getmedia/8b24119f-4f18-43e9-9d4f-6a3d08a21c59/AI-in-medezeggenschap-20251203.pdf

Named buyers and industry bodies

Crozier, R. (2026) ATO culture keeps a cap on agentic AI ambitions. iTnews, September 23. Available at: https://www.itnews.com.au/news/ato-culture-keeps-a-cap-on-agentic-ai-ambitions-629118

ISACA (2026) New ISACA Research: 56% of Digital Trust Pros Don’t Know How Fast They Could Shut Down AI After a Security Incident. Press release via Business Wire, March 23. Available at: https://www.businesswire.com/news/home/20260323090694/en/New-ISACA-Research-56-of-Digital-Trust-Pros-Dont-Know-How-Fast-They-Could-Shut-Down-AI-After-a-Security-Incident

↑ Back to the section guide

Disclosures: Independence, Financial, Intended Use, Methodology, Authoring, And Copyright

The evidence in this guide is current through October 15, 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 Consilium shared this guide with Oracle for review before publication. Editorial control, the analysis and the recommendations remained Greyhound Consilium’s own. 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 guide is published under the Greyhound Consilium brand and was developed in partnership with Oracle.

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). Greyhound Research holds no other financial interest in Oracle or its subsidiaries.

Intended Use

This guide is written for CIOs and the leaders who own finance, HR, supply chain and sales processes in organizations that run, or are moving to, Oracle Fusion Cloud Applications. It sets out practical steps for deciding how much authority to hand Fusion Claw, outcome by outcome, to inform technology, process and governance decisions. Its summaries of regulation are for business planning, not legal advice or a determination of any organization’s obligations. Greyhound Consilium is not a law firm, and readers should confirm specific obligations with qualified counsel. This guide is not investment advice, a recommendation to buy, sell, or hold any security, or a view on any company’s share price or valuation. Its guidance rests on the documents listed in Sources And References as they stood on October 15, 2026, and can change as Oracle and others publish more. Product availability, features and contract terms are described as Oracle published them through October 15, 2026, so confirm them with your Oracle account team before acting on them.

Research Methodology

Greyhound Consilium’s guidance rests on source evidence verified against the document that supports each claim. Greyhound Consilium developed this guide in partnership with Oracle in October 2026. It builds on Oracle Fusion Claw – Autonomy Arrives With A Receipt, the independent research report Greyhound Research published on October 14, 2026, and on Oracle’s public material published through October 15, 2026. The evidence comprises the thirty-seven sources listed in Sources And References, in eight classes, from Oracle’s announcements, documentation, blogs and executives’ statements in named media to the research this guide builds on, regulators’ texts, employee representatives and named buyers.

Every example is Oracle’s own published example or a named public source, linked where it appears. Exhibits carried forward from the research report say so on their face. Buyer evidence comes from Greyhound CIO Pulse 2026 and Greyhound Fieldnotes, as cited in the research report. No composite or invented customer appears, and the tools leave the figures to the reader. Quotations are checked against the live source before publication, and the guide was shared with Oracle before it was published.

Authoring Disclosure

This material reflects the analysis, viewpoint and editorial judgment of Greyhound Consilium. The position presented is developed from Greyhound Consilium’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 Consilium’s house style. AI does not independently establish Greyhound Consilium’s viewpoint, analytical position or conclusions. Material factual claims are verified against source evidence before release. Greyhound Consilium reviews and approves the final material and retains responsibility for it.

Copyright Policy

All content in this document is protected by copyright law and may not be reproduced, distributed, transmitted, displayed, published, or broadcast without the prior written permission of Greyhound Consilium 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 Consilium content. Quotation of up to one hundred words is permitted where it carries attribution to Greyhound Consilium 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. Oracle and its partners may share this guide with customers and at events, in full and unaltered. The published version of this guide on greyhoundresearch.com carries sharing tools for readers who wish to share it. Copyright in this guide is held by Greyhound Research, the legal entity under which Greyhound Consilium publishes.

For questions arising from these disclosures or about this guide, including licensing requests and media inquiries, contact Greyhound Consilium at connect@thofgr.com.

↑ Back to the section guide