Banks that use generative AI to answer customers now face a question from federal examiners that many of those systems struggle to answer: Where did that answer come from? That is the central finding of a new Inbenta analysis of the OCC’s supervisory framework for AI, announced today. Read the announcement.
The Office of the Comptroller of the Currency issued the framework on November 4, 2026, as a guide for how its examiners review the way banks use AI. Customer service is one of the uses it covers. Inbenta’s analysis finds that this is where banks are most exposed.
The framework warns that AI hallucinations about fees, rates, account status or policies could expose banks to legal liability and harm customers. It also makes clear that banks are still accountable, even when a vendor builds or hosts the AI. That leaves the bank to explain where each answer came from. Inbenta’s analysis shows how banks can do that before an examiner asks.
“Banks don’t need to pull back from AI. They need AI they can explain, including the source, the approval and the person who signed off,” said Melissa Solis, CEO of Inbenta.
The analysis covers:
- Where the framework’s hardest questions land for customer-facing AI: hallucinations, consistency and traceability
- The seven vendor-risk questions banks should put to any AI vendor, Inbenta included
- How Inbenta Encore answers them: subject matter experts approve answers at build time, that approval becomes the audit record, and each live answer is matched to approved content with its source attached
Read the full announcement and download Inbenta’s vendor-risk checklist.