Oracle announced a plan on October 26 to enforce AI access rules inside its databases and application infrastructure. The Deep Trust AI Architecture, unveiled at Oracle AI World in Las Vegas, combines database and application controls.
The company's announcement describes a blueprint combining Oracle AI Database, Autonomous AI Database and APEX AI Application Generator. Its central proposal is to make access restrictions apply even when an AI agent reaches a database directly.
Oracle says Deep Data Security checks a user's identity through a signed corporate-directory token and enforces access rules on each SQL request. SQL Assertions enforce business rules, such as an employee's spending limit. Deep Trust Data APIs restrict reads and writes to business objects needed for a task, such as purchase orders.
For application development, Oracle says its APEX generator produces declarative definitions that run in a shared execution engine. The company argues that this reduces the amount of AI-generated execution code organizations must secure and maintain. Those are Oracle's design claims; the announcement is not an independent security evaluation.
Permissions remain a separate problem
The National Institute of Standards and Technology identified agent identity and authorization as a security problem in a February 5 concept-paper announcement. Giving agents access to datasets, tools and applications creates risks that require appropriate identification and authorization controls, NIST said.
Its proposed project sought input on auditing, agent authority and defenses against prompt injection. That earlier work provides context for Oracle's announcement, not an endorsement of its products.
For businesses evaluating the blueprint, a concrete distinction is whether an agent can request an action and whether the underlying system permits it. Database enforcement can reject an unauthorized request even if an application submits it. That does not establish that every permitted request is accurate or appropriate.
