Oracle launches Deep Trust AI Architecture to secure enterprise systems

A blueprint for architectural trust

Oracle recently unveiled its Deep Trust AI Architecture, a framework designed to make security a built-in property of a company's technology stack. Instead of requiring every individual AI agent or application to correctly implement its own safeguards, this architecture places those protections deep within the system. This approach is intended to help organizations adopt AI more quickly while minimizing the risks associated with sensitive data.

The architecture is built upon four critical layers: the software platform, the data itself, data access, and application development. By embedding trust at these foundational levels, Oracle aims to ensure that security and business rules cannot be bypassed by AI models. This mirrors Oracle's Maximum Availability Architecture, which provides a blueprint for keeping mission-critical systems running reliably.

Juan Loaiza, executive vice president of AI Database Technologies at Oracle, noted that trust is currently the biggest challenge for enterprise AI. He explained that the new architecture allows organizations to put AI to work while maintaining control over the information and actions that AI-generated applications can take. The goal is to move the focus from just having intelligent models to having a trustworthy infrastructure to run them on.

Automated security and patching

One of the core components of this architecture is the Oracle Autonomous AI Database, which handles security updates automatically. This system applies patches to the database engine, operating system, storage, and firmware before they are made public. This proactive approach helps protect against attackers who use AI to find and reverse engineer vulnerabilities at high speeds.

A key feature of this process is rolling patching, which allows updates to proceed in stages. This means that applications can remain running and available to users without any disruption while the underlying system is being secured. Additionally, the database provides always-on encryption for data both at rest and in motion, featuring transparent decryption for authorized users.

Oracle is making these security and patching tools available to customers on co-managed and on-premises systems at a discount or for free through February 2027. The Autonomous AI Database is available across major hyperscalers as well as on-premises through Oracle Cloud@Customer. These tools are designed to eliminate the manual burden of keeping database systems secure as AI adoption expands.

Enforcing rules at the data source

A significant risk in AI deployment is that agents accessing databases directly might bypass the security rules usually found in application code. To solve this, Oracle Deep Data Security allows organizations to define end-user access rules directly within the Oracle AI Database. These rules are enforced on every SQL request, ensuring that the database itself verifies the user's identity via a cryptographically signed token from a corporate directory.

Complementing this is Oracle SQL Assertions, which are used to enforce specific business rules at the database level. For example, these assertions can prevent a transaction from occurring if a charge exceeds an employee's spending limit. Because these rules live in the database, they remain effective regardless of which application, end-user, or AI agent attempts to submit the change.

Furthermore, Oracle Deep Trust Data APIs provide controlled access to logical business objects, such as purchase orders. Rather than allowing agents to run arbitrary SQL commands, these APIs limit the AI to only the specific objects needed for a task. This ensures that the database maintains strict control over what the AI can see and modify during its operations.

Contextual understanding and generation

To help AI interpret business questions more accurately, Oracle introduced Deep NL2SQL. This tool provides the AI with necessary context, combining an organization's specific terminology and metrics with data knowledge about how information is organized. This helps the AI generate SQL queries that reflect the actual practices of the business rather than just generic data structures.

The system also includes Continuous Feedback Learning, which allows human users to correct inaccurate responses provided by the AI. These corrections are then validated through Knowledge Change Verification before they are incorporated into the corporate knowledge base. This cycle ensures that the AI's understanding of the business evolves accurately over time based on human input.

For building new applications, the Oracle APEX AI Application Generator allows coding agents to translate natural-language descriptions into APEXlang. This is an open, declarative application definition language used by the Oracle APEX runtime engine. Because the AI generates a definition rather than the actual execution code, organizations can build apps without having to secure and maintain massive amounts of AI-generated code.

Scaling AI without scaling risk

The overarching goal of the Deep Trust AI Architecture is to allow enterprises to scale their AI initiatives without a linear increase in risk. By providing a unified approach to how AI accesses, understands, and acts on data, Oracle aims to create a foundation for the 'agentic enterprise.' This means that every new AI agent or application created by a company automatically inherits the security and privacy protections of the underlying platform.

Industry analysts suggest that the competitive landscape for enterprise AI is shifting from the intelligence of the models to the trust of the architecture. While many companies are still trying to figure out how to govern AI agents, Oracle is attempting to address these issues at the source. By unifying intelligence, security, and governance, the architecture seeks to empower organizations to innovate faster.

The architecture provides a blueprint for putting AI to work on the most important data a company owns. By ensuring that security, privacy, and business rules are enforced by the platform and the runtime, companies can deliver more trusted outcomes. This framework is designed to help organizations move past the hurdles of AI trust to achieve practical productivity benefits.