Regulatory  ·  2026-11-18

OCC issues Supervisory Framework on Bank Use of Super Intelligence (OCC Bulletin 2026-54) for examiners

RegulatoryMedium impactUnited States
Following Executive Order 14434's use of the term 'Super Intelligence' (SI) for AI, the Office of the Comptroller of the Currency issued OCC Bulletin 2026-54, 'Super Intelligence: Supervisory Framework' (issued 4 November 2026), its first AI-specific examiner framework. It sets out 16 components under six pillars (purpose/permitted use, risk classification, controls/safeguards, governance/human oversight, operational resilience, testing/assurance) and an illustrative four-tier risk methodology (Tier 1 Critical e.g. autonomous payment agents / real-time credit decisioning; Tier 2 High e.g. AI-assisted underwriting, agent-initiated customer communications; Tier 4 Low e.g. internal productivity tools). It states SI should not be treated as a separate risk category and that banks are not required to set a distinct SI risk appetite. It explicitly says the document is an examiner resource, not new enforceable supervisory expectations, but flags hallucination risks on fees, rates, account status and policies and discusses customer remediation after an SI 'misfire'.
This is the OCC's clearest signal of how examiners will evaluate generative and agentic AI across national banks, federal savings associations and the US branches/agencies of foreign banks (relevant to international groups with a US presence). It previews exam questions on agent autonomy, reversibility, consequences and human oversight under existing standards, and treats inaccurate SI outputs (e.g. hallucinated fees/rates) as a consumer-compliance/UDAAP exposure.
Banks should inventory and tier generative/agentic AI use cases, document purpose and permitted use, define meaningful human-in-the-loop/human-on-the-loop oversight, and build controls to detect and contain agentic misfires and hallucinated fee/rate outputs.
PwC Our Take - OCC AI supervision (Nov 13)Consumer Finance Insights - OCC Super Intelligence framework
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →