A woman stands before glowing red steps curving through an illuminated arch in a dark gallery.

Oracle HCM

Fusion Claw for HR

Oracle's agentic applications for HR are moving from suggesting work to doing it. Fusion Claw is what runs them. This page explains what it is, the controls you set, and what to agree before you let it act on its own.

GuideUpdated 13 Nov 2026Fusion Claw

01 What Fusion Claw is

Oracle describes Fusion Claw as "a governed agentic execution runtime" for Oracle Fusion Agentic Applications. It was announced on 29 September 2026.

It helps to separate three terms:

An AI agent does a defined task, such as answering a policy question or drafting a summary. A person decides what happens next.
An agentic application works towards an outcome. It brings several agents, data sources and steps together.
Fusion Claw is what some of those applications run on. It lets them take on longer, larger pieces of work, replan as things change and, if you allow it, carry out the result.

Not every agentic application runs on Fusion Claw. On my Agentic Applications page, the ones that do are marked.

What this means for your team: Fusion Claw isn't something you switch on by itself. It arrives with the agentic applications that use it, so the decisions you make are about each application and how far you let it go.

02 How it works

Oracle splits the work in two:

A frontier model reasons and plans. It works out how to reach the outcome, and adapts the plan as conditions change.
Standard enterprise computation does the heavy work. Calculations and transactions at scale run as ordinary, repeatable software, not through the model.

Oracle says this means the model is only used where reasoning is needed. At launch, Oracle named Gemini and OpenAI models, with more to follow.

What this means for your team: the figures in a plan, such as costs and hours, come from calculation, not from the model guessing. The choices in the plan, such as which option to recommend, come from the model. That's where your review should focus.

03 The three controls you set

Oracle describes three parts that govern every run.

Enterprise Operating Envelope

Oracle says it captures your objectives, procedures, policies, constraints, permissions, risk thresholds, decision rights, approval requirements and escalation boundaries.

What to do: treat it as a written policy, not a settings screen. HR, payroll, legal and IT should agree it together, and someone should own it.

Outcome Trust Harness

Oracle says it enforces the operating envelope across every outcome, applying authority and controls to identity, capabilities, data and actions on each run.

What to do: check what each application can see and change. It should never see more than the people it works for.

Outcome Receipt

Oracle describes it as "an auditable record of exactly what happened, including the authority applied, evidence used, decisions made, actions and transactions executed, and the result."

What to do: decide where receipts are kept and who reads them. Keep payroll receipts with your payroll audit trail, and review a sample every quarter.

04 How much it does on its own

Oracle says customers choose how automated each process should be, "from quick assistance to governed full auto execution within explicitly delegated authority".

In practice, think of three levels:

Level	What happens	Good for
Review every plan	It prepares, a person approves, changes or rejects each one	Anything new, and anything that decides about a person
Approve within limits	It acts on routine cases inside agreed limits, and sends everything else for review	Repeatable work that specialists can check
Full auto	It acts within the authority you've delegated, and you review the receipts	Only where a wrong result is easy to spot and fix

Oracle's 26D AI Agent Studio guide puts it the same way: depending on the autonomy you allow, Fusion Claw "can recommend actions for your approval, perform tasks with your guidance, or complete tasks automatically within defined limits". It also says that before anyone can use it, "an application designer must configure an autonomous outcome".

My view: start every application at review every plan. Move up one level only after a few cycles where reviewers are genuinely changing or rejecting things, and the receipts show why. If nobody ever rejects anything, check whether anyone is really reviewing.

Work that decides who is hired, how they're paid or who is redeployed should stay with a person deciding each case, however good the recommendations look. Oracle's own HR announcement says "organizations retain responsibility for employment decisions".

05 The eight HR apps Oracle has announced on Fusion Claw

These are the apps named in Oracle's September and October announcements. Each name links to its entry on my Agentic Applications for HR page. Risk ratings use the same method as my AI HR Lifecycle Map.

App	Who it's for	What it does	My risk rating
Workforce Staffing	HR leaders	Builds staffing plans that meet coverage and service needs, balancing skills, schedules, labour rules and costs, and replans as conditions change	Take care
Workforce Orchestration	HR leaders	Builds workforce plans that weigh development, redeployment, hiring and external talent, then coordinates the actions approved through your review	Higher risk
Positions Command Center	HR specialists	Flags conflicts and downstream risks from position changes, and routes suggested corrections for review	Take care
Candidate Sourcing	Recruiters	Groups candidates and coordinates outreach, campaigns and follow-up within your recruiting processes and approvals	Higher risk
Recruiting Command Center	Recruiters	Brings hiring data together, highlights candidates ready to move forward, and prepares next steps for recruiters to approve	Higher risk
Workforce Compensation	Payroll and reward	Develops compensation recommendations against your criteria, budgets and pay guidelines, and prepares plans for review	Higher risk
Payroll Command Center	Payroll and reward	Flags likely payroll problems and their causes, and prepares follow-up actions for review	Take care
Payroll Activity Workspace	Payroll and reward	Compares current pay, balances and costs with earlier periods, flags significant differences, and prepares follow-up actions for review	Take care

Workforce Staffing was announced with Fusion Claw on 29 September 2026. The other seven were announced on 21 October 2026.

Where to start: the payroll and positions apps. Their output is checked by specialists who know what a correct result looks like, and a wrong flag costs time, not someone's job.

06 Fusion Claw in your 26D update

Oracle's announcements and its 26D release notes don't line up exactly. Oracle's list of AI features marks three 26D entries for HR as Fusion Claw:

In 26D	What Oracle's note says	What it means
Workforce Orchestration Command Center	Helps leaders and operations roles "identify workforce conditions that need attention and prepare evidence-backed follow-up actions"	The first delivered version is read-only. Oracle says it "summarizes, analyzes, and recommends" and doesn't make HCM changes. See my page on it.
Workforce schedule balance simulations, in Workforce Operations Command Center	Helps schedule managers compare scenarios when a schedule doesn't meet fill-rate, budget or policy goals, "while preserving hard scheduling constraints"	Claw-powered outcomes inside an app you may already use. Each scenario comes with a recommended plan to review before applying it.
Hiring questions and candidate recommendations, in Hiring Workspace for Store Managers	Retail hiring managers ask questions and get outcomes showing hiring risks, follow-up actions and candidate recommendations, with the information behind them	Oracle says hiring managers "remain responsible" for applying your hiring criteria and approvals before acting.

Workforce Staffing isn't in Oracle's 26D list of AI features yet, and nor are the other six apps announced in October.

What this means for your team: Fusion Claw may reach you first as new outcomes inside apps you already have, not as a new app. Check your 26D notes for these three before you plan anything around the announcements.

07 Agree these in writing first

Before any Fusion Claw app acts on its own, agree and write down:

The outcome. What the application is for, in one sentence.
The level. Which of the three levels it starts at, and what has to be true before it moves up.
The limits. Amounts, numbers of people, dates or grades it can't go beyond without a person.
The data. What it can see, and confirmation that it's no more than the people it works for can see.
The reviewers. Who approves, who can reject, and how a rejection is recorded.
The receipts. Where Outcome Receipts are kept, who samples them, and how often.
The stop. Who can pause it, and how quickly.
The update check. Who retests it after each quarterly update.

08 What Oracle hasn't said yet

Cost. Oracle's announcements don't mention pricing. SiliconANGLE reported that Fusion Claw is included with the agentic applications product, which is bought separately from a standard Fusion subscription, and that customers pay for AI units used. Confirm with Oracle before you plan budgets.
Dates for most apps. Of the eight announced apps, only Workforce Orchestration has a 26D release note so far, as Workforce Orchestration Command Center. For the other seven, wait for Oracle's readiness notes before you plan.
Evidence from live use. Oracle hasn't published measured savings or results from customers using these apps in production.

I'll update this page as Oracle publishes more.