About

OCC Issues Examiner Framework (Bulletin 2026-54) for Banks' Generative and Agentic AI Use

Published
Score
37

Why it matters

The OCC issued Bulletin 2026-54, "Super Intelligence: Supervisory Framework," on November 4, 2026. The non-binding examiner guide explains how national bank examiners will assess banks' use of generative and agentic AI in credit decisioning, payments, fraud detection, and customer service. It uses the term "Super Intelligence" (SI), which the OCC adopted to follow an executive order that replaced "artificial intelligence" in federal usage. The framework does not create a separate SI risk category. It applies existing categories, including governance, model risk, compliance, cybersecurity, and operational resilience. It sorts AI systems into tiers based on the consequences of failure, weighing autonomy, reversibility, and financial or customer impact. It also names agent-specific risks such as hallucinations about fees or account status, prompt injection, data poisoning, and scope drift. Its 16 supervisory components fall into six pillars.

The bulletin creates no new supervisory expectations or enforceable standards, and the OCC says banks are not expected to use it for self-assessment. The agency will revise it as the technology and supervisory experience change. How examiners will apply the risk tiers in actual examinations has not been shown yet. Commentary on oversight of vendor-operated agents, incident notification, and contractual liability for model errors comes from practitioners, not from the OCC.

The April 2026 revision, OCC Bulletin 2026-13, replaced SR 11-7 but excluded generative and agentic AI from its scope. Bulletin 2026-54 says the 2026-13 principles still govern non-generative, non-agentic models, and the new framework covers generative and agentic systems. Until now, those systems had no formal federal guidance. Although the bulletin does not bind banks, it signals how examiners will judge AI controls, human oversight, and vendor contracts. Banks should be reviewing AI inventories, use-case risk tiers, and vendor terms now. The OCC also said it expects bank activities to remain safe, sound, and lawful whether or not they use new technology.

Sources

mail Subscribe to Artificial Intelligence email updates

Primary sources. No fluff. Straight to your inbox.

Also on LawSnap