AI in action · Regulation
The OCC Told Examiners to Judge AI by What It Can Break
On November 4 the OCC issued its first guidance on how examiners should review banks' AI, telling them to weigh the consequences of mistakes rather than how sophisticated a model is.
America's largest national banks now know how their supervisor will look at artificial intelligence. On Wednesday, November 4, the Office of the Comptroller of the Currency issued what it described as its first-ever guidance for examiners on reviewing banks' use of AI, Reuters reported. The central instruction is a shift in emphasis: examiners should focus on the consequences of a potential AI-driven mistake, not on how advanced the model behind it is.
The document gives an example that cuts against intuition. A simple AI agent authorised to initiate transactions may pose more risk than a sophisticated model used only for internal analysis, because the first can move money and the second cannot. For higher-risk uses the OCC says banks may need tighter limits on autonomy or more direct human oversight, and examiners should weigh how a lender uses AI alongside the bank's size and complexity.
Equally important is what the guidance does not ask for. Examiners are told not to treat AI as a standalone risk category, and banks are not expected to set separate risk limits for it; AI is to be assessed inside the existing frameworks for operational, compliance and model risk. The guidance is not legally binding like a rule, but banks typically follow such documents closely, because they describe what examiners will check.
It also fills a gap the OCC left open itself. When the federal banking agencies replaced their model-risk guidance in April with OCC Bulletin 2026-13, they stated that generative and agentic AI models were "not within the scope" of that document because they are novel and changing quickly, and said separate guidance would follow. The new examiner guidance is the first piece of that, and it follows the Financial Stability Board's October report on sound practices for AI adoption, which asked for stronger human oversight of higher-risk and more autonomous systems.
For bank customers the effect will be gradual. Supervisors judging AI by its potential to cause harm will look hardest at systems that approve credit, move payments or talk to customers, and less at internal tools that summarise documents. That is also where banks have been most cautious, and the guidance gives them a clear reason to keep a human in the loop where an error would reach a customer's account.
Sources: Reuters (via WBOW) — US bank regulator issues first-ever AI guidance for examiners (Nov 4, 2026); Analytics Insight — US Banking Regulator Issues First AI Guidance for Examiners (Nov 2026); American Banker — Regulators' guidance on model risk leaves many questions unanswered