Oracle Plans ChatGPT Access to Fusion Apps

Oracle says employees will be able to run work in its Fusion business software from AI tools such as ChatGPT, through a new layer it calls Fusion Companion. The company announced the feature, along with a new in-app home screen called Fusion Agentic Home, at its AI World event in Las Vegas on 26 October.

The same day, Oracle set out a Deep Trust AI Architecture meant to stop AI agents from breaking security and business rules. Together the two announcements describe a bet: people may start their work in someone else’s chatbot, but Oracle wants the rules and the record to stay in its own software. Oracle gave no availability date or pricing for either Fusion feature.

In brief

Fusion Companion is meant to let users describe a goal in a supported AI tool such as ChatGPT and have Fusion pick the transactions, approvals and agents needed.
Deep Trust AI Architecture groups Oracle database and development tools into four layers that enforce rules for every agent and app.
Oracle is also offering security and patching tools free or at a large discount until the end of February 2027.

Fusion Companion lets a chatbot set the goal while Fusion does the work

Oracle Fusion Cloud Applications cover finance, HR, supply chain, manufacturing, sales, service and marketing. Oracle says Fusion Companion “automatically selects and coordinates the appropriate transactions, workflows, approvals, AI agents, and agentic applications” needed to finish a task, using the user’s own permissions. A user would no longer need to know which screen or workflow to open.

Users will reach it through supported AI tools such as ChatGPT, or through customer- or partner-built interfaces, Oracle says. It adds that Companion can draw on data from other company systems and coordinate authorised actions there too, while keeping “enterprise execution and accountability within Oracle Fusion Applications”.

“AI is changing where enterprise work begins, but organizations need to maintain control over how that work gets done,” said Chris Leone, executive vice-president of applications development. Fusion Agentic Home brings the same approach inside Fusion: staff state an objective, hand it to agents and track progress, with decisions needing human judgement passed back to people.

Companion can also call Oracle’s Claw-powered agentic applications. Oracle first launched 25 of those on 29 September, and its 26 October round-up restates them, with examples such as Ledger for accounting exceptions and Procurement Spend for spotting off-contract buying.

Deep Trust moves AI safeguards into the database

The Deep Trust announcement starts from a problem Oracle states plainly. When access rules sit only in application code, AI-written apps may get them wrong, and agents that query a database directly, for example through an MCP server or natural-language-to-SQL tools, “may bypass them entirely”.

Oracle’s answer has four layers. In the software platform, Autonomous AI Database applies security patches across the database, operating system, storage and firmware “before the software updates are made public”, Oracle says. In the data layer, Deep Data Security checks each user’s identity from a signed token issued by the company directory and applies access rules to every SQL request, while SQL Assertions block changes that break business rules, such as a charge above an employee’s spending limit. For data access, Deep Trust Data APIs let agents work only with defined business objects such as purchase orders, and Deep NL2SQL adds company terminology to plain-English questions. For app building, the APEX AI Application Generator has coding agents such as Codex write application definitions rather than running code.

“AI trust is the biggest challenge we face today,” said Juan Loaiza, executive vice-president for AI database technologies.

Oracle is betting AI rules belong in its software, not the agent

The two announcements fit together. If staff begin tasks in ChatGPT, Oracle risks becoming a back-end that another company’s assistant calls. Its answer is to keep the permissions, approvals and audit trail in Fusion and the database, wherever the request starts. Workday made a similar move this month when it linked its Sana agents with Anthropic’s Claude, and ServiceNow’s AI Passport tackles the same worry by checking what an agent may do at run time.

Oracle’s version leans on the database, which is its strongest product. That is a credible design: a rule enforced on every SQL request is harder for a careless agent to skip than one written into each app. But Deep Trust is a blueprint “bringing together proven capabilities”, in Oracle’s words, and the release does not say which parts are new or when each can be used. The free and discounted patching tools are the most concrete offer, and they arrive in the month Oracle shipped 1,295 security patches in its quarterly update.

Fusion Companion is still a promise. Oracle names ChatGPT as an example of a supported tool but gives no date, price or customer, and does not say whether OpenAI is a partner. Earlier at the same event, Oracle pledged $1 billion over four years to its partner network, and partner-built interfaces are one of the routes Oracle names for reaching Companion.

The detail to watch is the availability notice. Until Oracle says when customers can switch Companion on, and in which AI tools, this is a direction rather than a product.