Oracle launches agentic AI tools and Deep Trust architecture

From assistance to execution

Oracle is expanding its enterprise software portfolio by introducing a new class of software known as agentic applications. Unlike standard AI tools that might only provide suggestions, these applications are designed to analyze options, prepare actions, and carry out approved work. They operate within specific company policies and controls to handle complex business processes at scale.

The company launched 25 of these applications, which are powered by a runtime called Fusion Claw. This runtime is specifically built for longer-running and more complicated tasks that require deterministic execution. It allows the software to handle replanning and maintain tighter governance over the actions taken by autonomous systems.

These tools cover a wide range of operational departments, including finance, human resources, and supply chain management. For example, in finance, the tools can review accounting entries and propose follow-up steps for human approval. In HR, the software can assess hiring and redeployment options to help shape broader workforce plans.

New interfaces for delegation

To help employees interact with these new capabilities, Oracle introduced two new AI-based interfaces: Fusion Companion and Fusion Agentic Home. Fusion Companion acts as an intelligent layer that uses business context and user permissions to determine which workflows or agents should be involved in a specific task. It can even work with information from external systems while respecting existing security protocols.

Fusion Agentic Home serves as a personalized interface where staff can state a specific objective and hand the work over to AI systems. Users can then track the status of that work through to completion rather than performing every step manually. This setup is intended to let people focus on high-level decisions while the software coordinates the necessary steps to reach the result.

Oracle describes these innovations as a way to change how people work with enterprise software. By defining objectives and setting the policies, humans retain accountability while the AI handles the heavy lifting. This approach aims to give employees more capacity to focus on tasks that matter most to the business.

The Deep Trust AI Architecture

As AI systems gain wider access to sensitive business data, Oracle launched the Deep Trust AI Architecture to manage the associated risks. This framework applies security, privacy, and business rules across four distinct layers of the technology stack. These layers include the software platform, the data itself, how data is accessed, and the application development process.

At the platform level, the Autonomous AI Database can automatically apply security updates to the engine, operating system, and storage before they are made public. This includes rolling patching, which allows these updates to be introduced while applications continue to run without interruption. These measures ensure that the underlying infrastructure remains secure as AI features are deployed.

At the data layer, Oracle is using Deep Data Security and SQL Assertions to enforce rules directly in the database. This method is designed to stop AI-generated applications or direct database tools from bypassing policy controls. By enforcing rules in the database rather than just in the application code, the company aims to create a more robust security perimeter for enterprise data.

Contextual queries and development

For data access, Oracle introduced Deep Trust Data APIs to limit AI systems to specific business objects instead of giving them unrestricted access. They also launched Deep NL2SQL, which adds business and data context to natural-language database queries. This tool includes feedback and verification features designed to improve the accuracy of the responses generated by AI.

In the application development layer, the APEX AI Application Generator allows AI coding systems to translate natural-language requests into application definitions. This is different from producing large volumes of executable code, which can be difficult to maintain. Instead, the runtime engine enforces core protections, which Oracle argues reduces the security risks associated with generated code.

Juan Loaiza, Executive Vice President of AI Database Technologies at Oracle, noted that trust is a major barrier to using AI in business. He stated, "AI trust is the biggest challenge we face today." By building trust into the foundation of the platform, Oracle hopes to enable organizations to put AI to work while maintaining control over the actions the agents take.