AI that inherits your Oracle security
- AI previews
- User says yes
- Oracle runs it as that user
- Audited
If a user can't see it or do it in Oracle, they can't do it through our AI. Approval is a step in the flow that the agent cannot route around.
Eight principles we build in
- Runs as the userReads and updates live data under the user's own role, responsibility or token.
- Approval before writesA person confirms each change. The agent cannot route around it.
- Inside your environmentRuns in your Fusion pod or your OCI tenancy or data centre. No extra data movement for Fusion apps.
- The AI never touches the databaseIt can only call a fixed set of approved tools.
- Queries validated by codeEvery generated SQL, ROQL or CDP query is checked against the real schema first.
- Sensitive data protectedBank, IBAN and tax identifiers are removed before anything reaches the AI on EBS.
- Full audit trailEvery query, action and AI call is logged with user, time and outcome.
- Fail-closedIf a check fails, nothing runs and no data is returned.
Your choice of AI model
Choose by budget and data policy: enterprise cloud AI such as Claude through the enterprise API (data not used for training), OCI Generative AI under your Oracle agreement, or an open model on your own servers so no data leaves your network. The model sees only the question and an already-filtered answer. It never connects to your systems.
Secure tools for any AI client
When we expose your integrations to AI through MCP, the same controls apply: trusted sign-in (SSO, OAuth 2.1 or mobile OTP), a tool policy for what can be read or written, approval on writes, and a full audit trail.
Questions about security?
Bring them to Oracle AI World. We will walk through the controls for your Oracle product.