APEX 26.2: your AI agent can't see more data than your user

Oracle APEX 26.2 was released on 6 October 2026, on-premises first, with Autonomous AI Database and OCI to follow. The feature I'd read first isn't in the UI. It's in the security layer.

What it does

• APEX passes the authenticated end-user identity (and the application identity) through to Oracle AI Database: the Deep Data Security integration.
• Declarative Data Roles and Data Grants (CREATE DATA ROLE, CREATE DATA GRANT) control access by row, column and individual value, enforced by the database kernel.
• SQL run by supported AI Agent tools uses the requesting user's security context, not one shared application account.
• Agents can send users to application pages, but you define the allowed destinations and APEX builds the URL and checksum, so session details stay out of the model's messages.
• Providers: a new Provider API setting (OpenAI Responses, Gemini Interactions, Responses for OCI Generative AI) and a configurable Reasoning Effort per agent or per request.

Takeaway: before you put an agent on top of your APEX app, move "who can see what" out of prompts and page conditions and into the database. A prompt is a request; a data grant is enforced at query time. Sources: Configuring Deep Data Security with APEX, APEX 26.2 new features, Announcing Oracle APEX 26.2.