Oracle unveils Deep Trust AI Architecture for enterprises

Oracle has introduced its Deep Trust AI Architecture, a framework for enterprise AI systems.

The architecture outlines how Oracle wants customers to manage security, privacy, business rules and application development as more AI agents and AI-generated software are used within existing technology estates.

Instead of leaving those controls to individual applications or agents, the model embeds them across four layers of the stack: software platform, data, data access and application development. Oracle compared the approach with its Maximum Availability Architecture blueprint, long used by customers to guide system resilience.

The announcement comes as large companies face growing concerns about how AI tools interact with sensitive records, apply internal policies and whether responses generated from corporate data can be trusted. Those concerns have intensified as natural language interfaces and autonomous agents gain broader access to enterprise systems.

At the software platform layer, Oracle is positioning its Autonomous AI Database as a way to limit exposure to newly discovered vulnerabilities. It said the product applies security updates automatically across the database engine, operating system, storage and firmware before those updates are made public, while rolling patching allows systems to remain online during the process.

The database also uses encryption for data at rest and in transit, while authorised users and applications can decrypt information transparently. Oracle said the service is available on major hyperscaler clouds as well as through Oracle Cloud@Customer for on-premises deployments.

Database controls

A central part of the new architecture is moving access controls and business rules into the database itself. Oracle said that if those rules remain only in application code, AI-generated software may implement them incorrectly, while direct database access by agents could bypass them.

To address that, Oracle is using Oracle Deep Data Security, which it said lets organisations define end-user access rules directly in Oracle AI Database and apply them to every SQL request. The database verifies a user's identity from a cryptographically signed token issued by a corporate directory, then determines what that user is allowed to access.

Oracle is also promoting Oracle SQL Assertions as a way to enforce business rules at source. It said these checks can stop invalid actions, such as charges above an employee's spending limit, regardless of whether the request comes from an application, a person or an AI agent.

Data access

Another element is a controlled API layer that limits what AI systems can read or write. Oracle said its Deep Trust Data APIs provide access to logical business objects such as purchase orders, rather than allowing unrestricted SQL queries against databases.

The approach is intended to narrow the scope of what an agent or application can do in a given task, while leaving the database to enforce rules over what can be seen or changed. It reflects a broader industry move towards restricting AI systems to tightly defined interfaces rather than open-ended access to operational data.

For natural language database queries, Oracle is also introducing Deep NL2SQL. It said the system combines business context, including company terminology, practices and metrics, with information about how data is defined and organised, so AI can generate SQL that better reflects the business setting.

Oracle said the feature also includes Continuous Feedback Learning, which allows users to correct inaccurate responses, and Knowledge Change Verification, which is intended to validate those corrections before they are added to the corporate knowledge base.

Application generation

At the application layer, Oracle is linking the new architecture to Oracle APEX AI Application Generator. The tool lets AI coding agents convert natural-language prompts into APEXlang, which Oracle describes as an open declarative application definition language.

Oracle said Oracle APEX then executes those definitions in a central runtime engine. The company's argument is that customers do not need to trust and maintain large volumes of AI-generated execution code if the AI only creates an application definition and the runtime handles execution.

Developers can use VS Code, Oracle APEX or other AI coding agents, including Codex, according to Oracle. It said this model can simplify maintenance and updates because the execution environment remains standardised.

Juan Loaiza, executive vice president, AI database technologies, Oracle, said the challenge for companies is confidence in AI outputs and actions. "AI models can now enable amazing productivity and innovation. However, enterprises can't realize these benefits if they can't trust what AI generates. AI trust is the biggest challenge we face today," Loaiza said.

He added: "With Oracle Deep Trust AI Architecture, we are building trust into the underlying software platform, data, data access, and application development layers. This enables organizations to put AI to work while maintaining control over the information and actions that AI agents and AI-generated applications can take."

Steven Dickens, CEO, HyperFRAME Research, said Oracle was trying to shift the focus of enterprise AI competition towards how trust is embedded in underlying systems. "The enterprise AI race is entering a new phase, and Oracle is shifting the competitive battleground from model intelligence to architectural trust," Dickens said. "While much of the industry is still grappling with securing AI agents and governing access to enterprise data, Oracle is addressing these challenges at the source. With Deep Data Security enforcing policies directly in the database and Deep NL2SQL bringing business context, continuous learning, and verification to AI-generated answers, Oracle is fundamentally changing the equation for enterprise AI. What sets the Deep Trust AI Architecture apart is its unified approach to embedding trust into how AI accesses, understands, and acts on enterprise data. Oracle is raising the industry bar by unifying intelligence, security, and governance at the foundation of the agentic enterprise, empowering organizations to innovate faster, scale AI with confidence, and deliver more trusted business outcomes."