Accenture builds agentic system with Oracle AI Database to protect client data
Oracle Autonomous AI Database 26ai with Oracle Deep Data Security helps isolate client information as agents generate SQL to enhance delivery lifecycle.
Ireland | Professional Services
“Oracle Deep Data Security does not expose our client data to people who are not authorized to see it. This increases trust with our clients for a competitive advantage.”
Firm seeks to build a knowledge base for the technology delivery lifecycle
Accenture’s global consulting practice set out to build an enterprise-class, multitenant knowledge base—its Advanced Insights Engine—to power agentic applications across the full technology delivery lifecycle (TDLC). Rather than appoint tools for a single activity, the goal was to build one governed platform that hundreds of users could use from end to end: to win work faster, learn from every pursuit, and deliver with higher quality and speed once an engagement begins.
Early sales and requests for proposals (RFP) response development served as the first proving ground; for example, agents can assemble an automated RFP response, including architecture diagrams, from a highly curated document corpus.
Accenture built the application in a few days on Oracle Cloud Infrastructure (OCI) with Oracle Autonomous AI Database 26ai and Oracle Deep Data Security. From the beginning, the architecture had a firm requirement: SQL generated by an AI agent or Model Context Protocol (MCP) server could not become a path around client data isolation.
Agent-generated SQL changes the security requirement
Protecting data based on identity and business context is not a new problem. Accenture’s approach reflects decades of experience with database security, from application views that filtered rows based on context to Oracle Virtual Private Database, which added standardized row-level security.
Agentic applications raised a different concern. Oracle Virtual Private Database is a mature capability, but Accenture wanted authorization that did not depend on every application or MCP component correctly applying programmatic security logic.
That distinction mattered because the platform is designed to contain client information. Accenture wanted a control at the data layer that would continue to apply when an agent generated SQL dynamically or when an upstream component did not behave as expected.
Oracle Deep Data Security puts authorization with the data
Accenture selected Oracle Deep Data Security because authorization policies are declarative and enforced in the database rather than implemented as application-specific code. For the team, that represented the same kind of shift that database-enforced relational integrity once brought to application development: Rules can be defined once and applied where the data resides.
The distinction from earlier approaches is important to the architecture. Instead of relying on application code to add the appropriate access logic for every query, Oracle Deep Data Security applies authorization as SQL is processed by the database. This enables the Accenture team to apply engagement-specific controls to client information within the same multitenant platform.
Security supports the full delivery lifecycle—at scale
The knowledge base now supports dozens of agentic applications and well over 100 people across Accenture’s delivery organization. Those applications span the whole lifecycle—from RFP response and win-loss evaluation through requirements and fit-gap analysis and the generation of prototypes, designs, code, tests, and documentation, and finally to Hyper Care after go-live. Accenture also demonstrated that the platform can scale for 3X concurrent demand—300 simultaneous sessions with query latency of 15 seconds or less for the majority of queries.
The earliest and most measurable gains came from RFP work: response time has dropped from weeks to days, with much higher confidence in quality because teams can draw from a large body of curated information—resulting in an estimated $75,000 to $80,000 saved per RFP response. The same curated foundation and agentic tooling now extend those gains across later phases, from fit-gap analysis to Hyper Care, helping teams bring value to clients sooner.
For Accenture, the larger lesson is that AI security cannot depend on one boundary. As agents gain access to more enterprise systems and sensitive information, the team is pairing perimeter, identity, application, and infrastructure protections with authorization where the information is stored.
That data-layer control gives Accenture a foundation for placing client information in a shared agentic platform while keeping engagement data isolated, without assuming every agent or upstream component will always behave as intended.