Oracle Corporation (NYSE: ORCL) has introduced Deep Trust AI Architecture, a framework intended to keep AI agents and AI-generated applications within enterprise data-access, security and business-rule controls. The approach moves those safeguards into core platform, data, data-access and application-development layers rather than relying on every individual agent or application to apply them correctly.
That distinction matters for organizations expanding the number of systems that can query databases, generate code or take actions on business data. When policies live only in application code, the company said, an AI-generated application can implement them incorrectly and an agent using a model-context-protocol server or natural-language-to-SQL interface may bypass them. Oracle’s design is intended to give IT teams a common control point below those tools.
At the software-platform layer, Oracle said its Autonomous AI Database automatically applies security updates across the database engine, operating system, storage and firmware before the updates are public. Its rolling-patching process is designed to update systems in stages while applications remain online. The database also uses always-on encryption for data at rest and in motion, according to the company.
The data layer centers on Oracle Deep Data Security and SQL Assertions. Deep Data Security lets an organization define end-user access rules directly in Oracle AI Database; the database uses a cryptographically signed token from the corporate directory to identify a user and determine which data that user may access. SQL Assertions apply business rules in the database itself. Oracle gave the example of blocking an expense charge that exceeds an employee’s authorized spending limit, regardless of whether a person, application or AI agent submitted it.
For agent and application access, Deep Trust Data APIs expose controlled read and write access to logical business objects, such as purchase orders, rather than granting arbitrary SQL access. Oracle said the database enforces the associated rules governing what a requester can see or change. This addresses a practical governance issue for IT teams: limiting an AI task to the objects it needs without opening unrestricted access to underlying tables.
The architecture also includes Deep NL2SQL for business users querying data in natural language. Oracle said the capability combines business terminology, practices and metrics with information about how corporate data is defined and organized. Its Continuous Feedback Learning and Knowledge Change Verification functions are intended to let users correct inaccurate answers and validate changes before they enter a corporate knowledge base.
On the application-development layer, Oracle APEX AI Application Generator turns natural-language descriptions into APEXlang, an open declarative application-definition language. Oracle APEX then executes those definitions through its centralized runtime engine. The company said this separates an AI system’s role in describing application functionality from the security, scalability and availability functions of the engine that runs it.
For security and data-platform teams, the value of the architecture is a blueprint for enforcing controls where agent behavior meets enterprise systems, rather than reconstructing policy checks in each new application. It does not remove the need to define access rules, validate generated work or govern identities. But by placing those checks at the platform and data layers, Oracle is targeting a limitation of legacy application-level controls: they protect data only when every request follows the expected application path.