Oracle on Monday introduced its Deep Trust AI Architecture at its AI World conference in Las Vegas. The approach builds security, privacy and business rules into the database layer so AI agents and AI-generated apps can't bypass them.
Why it matters: Companies want to let AI agents query and act on enterprise data, but rules written into application code can be skipped when an agent goes straight to the database.
The pieces
According to Oracle and a blog post by database chief Juan Loaiza and Maria Colgan, the architecture includes:
- Deep Data Security, which enforces per-user access rules in the database itself.
- SQL Assertions, which block data that breaks business rules, such as an expense over an employee's limit.
- Trusted Data APIs, which give agents access to business objects instead of unrestricted SQL.
- Deep NL2SQL, which adds business context and feedback loops to plain-English data questions.
- APEX AI App Generator, which turns descriptions into APEXlang, a declarative app language.
Oracle's pitch
"AI trust is the biggest challenge we face today," Loaiza said in the announcement. Oracle compared the blueprint to its Maximum Availability Architecture and said it runs on major cloud providers and on-premises.