US Bank Regulator OCC Issues Its First AI Guidance for Bank Examiners

The agency released the document as News Release 2026-97 and OCC Bulletin 2026-54. The OCC's own documents do not say "artificial intelligence". They use "Super Intelligence", or SI, and "agentic SI". According to PYMNTS, this follows a September 29 executive order from President Donald Trump that replaced "artificial intelligence" with "super intelligence" in executive-branch usage. Other reports on the framework, including Reuters, refer to it as AI guidance.

What the Framework Covers

The OCC said examiners will use the framework to assess how banks use AI and AI agents in their normal business, along with the risks and controls involved. According to the ABA Banking Journal, that includes credit decisions, payments, fraud detection, customer service, operations and risk management. The agency said it will update the framework regularly as the technology changes and as stakeholders give feedback.

The central instruction is that AI should not get its own risk category. The OCC said "examiners should not treat SI as a separate and distinct risk category", and banks will not have to set a separate risk appetite for it. Instead, examiners are to add AI-specific considerations to the existing work they already do on model risk management, governance, technology, compliance and operational risk.

Risk Based on Consequences

Reports on the framework say it tells examiners to focus on the harm an AI error could do, not on how technically advanced a model is. In the example reported, a simple AI agent that can initiate transactions could carry more risk than a sophisticated model used only internally. Analytics Insight reported that higher-risk uses may call for tighter limits on how much AI can act on its own, and for stronger human oversight where systems make important decisions or carry out financial activity. Examiners are also expected to take account of each bank's size, activities and risk profile, including at smaller community banks.

The OCC called its approach technology-neutral. It said it expects banks to operate safely, soundly and within the law whether they use traditional or novel technology.

Not a Binding Rule

The document does not set enforceable standards or new supervisory expectations for banks, according to PYMNTS, and it is not meant as a tool for banks to assess their own AI risk management. In practice, though, banks tend to follow examiner guidance closely because it shapes how they are examined.

The framework builds on earlier OCC work. In its spring 2026 Semiannual Risk Perspective, the agency described AI as both a risk and an opportunity for innovation, and it urged banks to strengthen defences against AI-enabled cyber threats. Banks have used machine learning for years in credit scoring and fraud detection, and many are now testing generative and agentic tools.

Why It Matters

For banks and the technology vendors that serve them, the guidance shows how US examiners will look at AI deployments in practice. Scrutiny will depend on what a system is allowed to do, not on how capable it is. That matters most for agentic AI. Banks that let software agents move money or make customer decisions should expect closer questions about controls and human oversight than those using AI for internal analysis.

The guidance comes as financial regulators in several jurisdictions, including the UAE, are deciding how to supervise AI in banking. Earlier on Wednesday, the Central Bank of the UAE listed thematic reviews of artificial intelligence among its supervisory priorities for 2027.