Accenture Demonstrates How Oracle's Deep Data Security Shifts Trust to the Database Layer
As AI agents increasingly handle application logic, traditional security models break down. Accenture's work with Oracle shows how database-enforced access policies can protect sensitive data regardless of what queries an AI system generates.
Oracle Corp.'s Deep Data Security rests on a fundamental observation about modern enterprise security: the moment an AI agent becomes the application layer, the traditional trust boundary no longer holds. Moving that boundary into the database itself offers a solution, one that Roger Cornejo, technology innovation principal director of gen AI/AI at Accenture Enkitec Group, has now validated through published research.
Working with Oracle's security platform, Cornejo's team constructed an application that demonstrates the approach in practice. The system allows the database to enforce access policies independent of any Structured Query Language an AI agent might construct. "We speak to a client, and we walk them through our security model, including the layered security," Cornejo said. "I can tell you, every person that we have shown this application to, [with] this level of security, has not seen anything like it. They can trust the application is not going to expose their data to people who aren't authorized to see it."
Cornejo presented these findings during Oracle's "AI Cyberattacks Are Escalating: How to Secure Your Data Now" event, speaking with Dave Vellante on theCUBE, SiliconANGLE Media's livestreaming platform.
Defining boundaries in enterprise security
Within Accenture's Oracle practice, the implementation centers on a Model Context Protocol server that grants AI agents access to a knowledge repository containing more than 925,000 documents distributed across 27 Oracle product namespaces. Each agent operates under different authorization levels, with Deep Data Security establishing boundaries directly in SQL. The database then enforces these boundaries for every agent on every query.
https://www.youtube.com/embed/CIIbV_VNV5Q?feature=oembed
"When we were tasked to build this application out, it needed to be multi-tenant, it needed to scale, it needed to be robust in terms of security," Cornejo explained to theCUBE. "We implemented this all with database security. We have dozens, probably 40 or more agents that are hitting this application. We needed to make sure it would scale as well. Right now, we have tested it to scale to 300 simultaneous queries returning the results, most of the results, in 15 seconds or less."