Oracle’s Deep Data Security Shifts Trust Boundaries for AI Applications
Oracle Corporation’s Deep Data Security platform introduces a novel approach to enterprise security, particularly as artificial intelligence (AI) becomes integrated within application layers. This paradigm shift necessitates repositioning the trust boundary away from traditional layers.
According to Roger Cornejo, technology innovation principal director of generative AI at Accenture Enkitec Group, the trust boundary can effectively be shifted to the database level. Cornejo shared insights from his research, illustrating how this approach works with Oracle’s Deep Data Security. His team developed an application that utilizes this security solution, capable of enforcing access policies irrespective of the SQL queries generated by the AI agent.
In a recent discussion with Dave Vellante during Oracle’s “AI Cyberattacks Are Escalating: How to Secure Your Data Now” event, Cornejo emphasized the unique level of security demonstrated through their application. “Clients have remarked that they have never encountered anything like this, assuring them that their data is shielded from unauthorized access,” he stated.
Establishing Security Boundaries in the AI Era
Cornejo’s role at Accenture involves implementing Oracle’s Model Context Protocol server, which grants AI agents access to a comprehensive knowledge repository containing over 925,000 documents across 27 Oracle product namespaces. This framework assigns varying levels of authorization to different agents, while Deep Data Security delineates boundaries within SQL queries—ensuring the database upholds these restrictions for each agent and query executed.
“Our design objective was to create an application that was not only multi-tenant but also scalable and robust in security,” Cornejo explained during his conversation with theCUBE. “We achieved this by leveraging database security, managing interactions from over 40 agents simultaneously. We have successfully tested the application to handle up to 300 concurrent queries, delivering results in under 15 seconds for the majority of requests.”
The discussion is part of theCUBE’s coverage of the pivotal Oracle event, where cybersecurity experts converge to tackle rising threats in the era of AI:
(Disclosure: TheCUBE serves as a paid media partner for Oracle’s “AI Cyberattacks Are Escalating: How to Secure Your Data Now” event. Editorial control over content lies exclusively with theCUBE and SiliconANGLE, independent of Oracle or other sponsors of the event.)