The OCC has no AI rule. Its main AI document is Bulletin 2026-54 of November 4, 2026, a framework for its examiners that mainly covers generative and agentic AI and says it sets no expectations for banks. Traditional models, including machine learning, fall under the model risk guidance in Bulletin 2026-13, and existing laws apply to AI as to anything else.
TRACKED DOCUMENTS: OCC Bulletin 2026-54 · OCC Bulletin 2026-13 · OCC Bulletin 2023-17
Main document
OCC Bulletin 2026-54, November 4, 2026
What it is
A resource for OCC examiners, not a rule for banks
Covers
Mainly generative and agentic AI
Shape
One principle, six pillars, sixteen components, four illustrative tiers
Deadline
None. The framework has no effective date
What is the OCC and which banks does it supervise?
The Office of the Comptroller of the Currency (OCC) is the federal agency that supervises national banks, federal savings associations, and federal branches and agencies of foreign banks. Its AI documents speak to those institutions and to its own examiners.
Supervising a bank means sending examiners to look at how it is run and whether it is safe. Examiners are OCC staff who review a bank's records, controls and decisions. In practice, if your bank has “National Association” or “N.A.” in its name, the OCC is very likely its federal supervisor. State banks are supervised by the Federal Reserve or the FDIC together with their state, so OCC documents do not speak to them directly. Our OCC page lists everything we track from the agency.
Does the OCC have an AI rule?
No. The OCC has issued no regulation on AI. What it has issued is guidance and, in November 2026, a framework written for its own examiners.
The three kinds of document carry different weight:
- A rule (regulation) is law. It goes through public notice and comment, and a bank that breaks it can face enforcement.
- Guidance describes sound practice. It shows how the agency thinks, but it is not law. Bulletin 2026-13, for example, says it does not set “enforceable standards or prescriptive requirements”.
- An examiner resource is written for the agency's own staff. Bulletin 2026-54 says its framework “does not establish supervisory expectations for banks” and that banks “are not expected to use this document to assess their use of SI”.
Put simply, existing laws still apply to anything a bank does with AI. The News Release 2026-97 says the OCC expects bank activities to be conducted in a safe and sound manner and consistent with applicable laws, whether they are traditional or involve new technology. The framework itself says in section 2.8 that AI does not change a bank's underlying legal obligations.
What has the OCC issued on AI, in date order?
Ten OCC documents bear on AI, from a 1997 bulletin on credit scoring to the November 2026 framework. Only third-party guidance and the 2026 model risk guidance are in force as guidance today, and none of them is a rule.
| Document | Date | What it is | Does it bind a bank? |
|---|---|---|---|
| Bulletin 1997-24 | May 20, 1997 | Examination guidance on credit scoring models. | No. Rescinded April 17, 2026 by Bulletin 2026-13. |
| Bulletin 2011-12 | April 4, 2011 | Model risk management guidance, issued with the Federal Reserve (its SR 11-7). | No. Rescinded April 17, 2026 by Bulletin 2026-13. |
| Bulletin 2021-17 | March 31, 2021 | Five agencies' joint request for information on how financial institutions use AI, including machine learning. | No. It asked questions. Bulletin 2026-13 says the agencies plan a new request for information on banks' use of AI. |
| Bulletin 2021-19 | April 9, 2021 | Interagency statement on model risk management for BSA/AML systems (the anti-money-laundering systems that monitor transactions). | No. Rescinded April 17, 2026 by Bulletin 2026-13. |
| Bulletin 2023-17 | June 6, 2023 | Interagency guidance on third-party relationships. A third party is any outside company the bank works with, including an AI vendor. | Guidance in force, not a rule. |
| Acting Comptroller Hood speech | April 29, 2025 | Recorded remarks on AI in financial services at the National Fair Housing Alliance's Responsible AI Symposium. | No. A speech states views. |
| Bulletin 2026-13 | April 17, 2026 | Revised interagency model risk management guidance. Places generative and agentic AI outside its scope. | Guidance in force. It says non-compliance “will not result in supervisory criticism”. |
| Semiannual Risk Perspective, Spring 2026 | May 7, 2026 | The OCC's risk report. Describes AI changing the cyber threat landscape and banks' measured approach to generative and agentic AI. | No. A report. |
| Bulletin 2026-46 | September 11, 2026 | Proposed interagency guidance to revise and replace the third-party risk guidance, with a request for comment. | No. A proposal. It would apply when final. |
| Bulletin 2026-54 | November 4, 2026 | The “Super Intelligence: Supervisory Framework” for examiners, mainly about generative and agentic AI. | No. It says it sets no expectations for banks. |
Each name links to our document page with the official link and status. Two documents matter most for AI today. Bulletin 2026-13 covers traditional models, including machine-learning ones. Bulletin 2026-54 covers the generative and agentic AI that 2026-13 leaves out. Footnote 1 of the framework draws that line itself. Generative AI writes new text, images or code. Agentic AI can take actions on its own, such as sending a payment or updating a record.
What does OCC Bulletin 2026-54 say, in one screen?
The framework says the controls around an AI system could grow stronger as the harm it could cause grows. It sorts its advice into six pillars, sixteen components and four illustrative risk tiers. The full reference guide walks through every section.
The overarching principle (section 1.1). A bank's “control intensity could increase with the potential consequences of an SI agent's failure or misuse”. Risk may be judged by consequences, by how much the system acts for the bank, and by human oversight, not by how advanced the technology is. The OCC's own example: a simple rules-based agent that can start transactions could be riskier than a sophisticated model that only helps draft internal text.
The six pillars (section 1.2).
- Purpose: what the system is allowed to do.
- Risk: what happens if its output or action is wrong.
- Control: what stops unsafe or unauthorized actions.
- Oversight: who watches the system and who answers for it.
- Resilience: whether the bank can detect, contain and recover from a failure and keep the business running.
- Assurance: evidence that the controls keep working.
The sixteen components (sections 2.1 to 2.16).
- 2.1 Defined purpose and permitted use: a written statement of what the system may and may not do.
- 2.2 Risk classification and criticality assessment: sorting each use by the harm it could cause.
- 2.3 Human-in-the-loop and human-on-the-loop controls: a person approves each action, or watches and can step in.
- 2.4 Governance and accountability: named owners, board information and approval points.
- 2.5 Model risk management extensions: adapting model checks to systems whose answers vary run to run.
- 2.6 Data governance: where data comes from and whether it fits the use.
- 2.7 Privacy: models that may memorize or infer personal data, and vendor use of bank data.
- 2.8 Compliance: BSA, sanctions, consumer rules and wrong answers about fees or rates.
- 2.9 Third-party and vendor risk: AI bought from outside, including a provider's own agents.
- 2.10 Operational resilience: a system that stays up while acting wrongly.
- 2.11 Cybersecurity: attacks such as prompt injection, where hostile text tricks the system.
- 2.12 Performance monitoring: watching live behavior for drift, meaning a slow change from how it was tested.
- 2.13 Testing: methods beyond one expected answer, including red teaming and hallucination checks.
- 2.14 Documentation and auditability: records good enough to rebuild a decision later.
- 2.15 Internal audit and independent challenge: assurance from people who did not build or check the system.
- 2.16 Reporting and transparency: timely reports on use cases, changes, incidents and customer harm.
A hallucination is a confident answer that is false. More on each component is in the sixteen framework components.
The four illustrative tiers (section 3.2).
- Tier 1 Critical: direct action that is hard to undo, with high impact. Example given: an autonomous payment agent.
- Tier 2 High: direct action or heavy decision support with a human watching. Example given: AML alert disposition.
- Tier 3 Moderate: decision support where a human must approve. Example given: summarizing documents for credit memos.
- Tier 4 Low: no customer or financial data and no direct action. Example given: general employee Q&A.
The OCC calls these tiers illustrative. See the four OCC AI risk tiers for the seven scoring factors in section 3.1.
What is binding on a bank today, and what is not?
Laws and regulations bind a bank whether or not AI is involved. The OCC's AI documents themselves do not: they are guidance, proposals, reports or an examiner resource.
| Binding | Not binding |
|---|---|
| Existing laws and regulations, such as consumer protection, fair lending and the Bank Secrecy Act (BSA, the anti-money-laundering law). Section 2.8 says AI does not change these obligations. | Bulletin 2026-54 and its framework. The bulletin says it sets no “enforceable standards or prescriptive requirements”. |
| Notifications that laws, regulations or contracts require. Section 2.16 says these could apply to AI use cases. | Bulletin 2026-13. It says non-compliance “will not result in supervisory criticism”. |
| The OCC's general expectation, in News Release 2026-97, of safe and sound operation consistent with applicable laws. | Bulletin 2023-17, the 2026-46 proposal, speeches and the Semiannual Risk Perspective. |
Two points are easy to miss. The framework says examiners should not treat AI as a separate risk category, and banks are not required to define a separate AI risk appetite (section 1). A risk appetite is the amount of risk a board agrees to accept. Instead, AI risk sits inside existing categories such as operational, compliance and credit risk. Fair lending, meaning lending without discrimination, is one place where the laws already apply. Our page on data, privacy and consumer compliance covers it.
Why does the OCC say “Super Intelligence” instead of AI?
The OCC writes “Super Intelligence (SI)” where everyone else writes AI. It means the same technology, renamed under an October 2026 executive order.
Footnote 2 of Bulletin 2026-54 says “Super Intelligence” refers to the technologies and systems previously covered by the term “Artificial Intelligence” as defined in 15 USC 9401(3), plus the “promise, potential, and rapidly advancing capabilities of those technologies”. It cites Executive Order 14434 of October 2, 2026. The lexicon in Appendix A says that definitions borrowed from an industry AI lexicon were changed from “AI” to “SI” for consistency.
Put simply, “SI” in an OCC document is not a claim about machines smarter than people. It is the same chatbots, models and agents banks already use. This series writes “AI”. Footnote 3 of the framework also says it was prepared with the assistance of AI and then reviewed, edited and approved by a human.
Where should I start, depending on my role?
Start with the page that matches your job. Each one explains a part of the framework in plain words and names the section it comes from.
| If you are | Read first | Then |
|---|---|---|
| A board member | Board, management, audit and reporting | The four risk tiers |
| In compliance or risk | Data, privacy and consumer compliance | Preparing for an OCC exam |
| In technology or model validation | Model risk management for generative AI | Resilience, kill switches and cybersecurity |
| At a community bank | What it means for community banks | Vendor and third-party AI |
| A vendor selling AI to banks | Vendor and third-party AI | Human oversight of AI agents |
Validation means an independent check that a model works as intended. A kill switch is a way to stop an AI system at once. Words you do not know are in the OCC AI glossary.
What pages are in this OCC AI guidance series?
The series has a hub (this page), the full reference to Bulletin 2026-54, and one page per part of the framework. They are listed in reading order.
- OCC AI guidance in plain English: Everything the OCC has issued on AI, in order, what binds a bank and what does not, and where to start.
- OCC Bulletin 2026-54, section by section: The full reference to the November 2026 examiner framework for generative and agentic AI.
- The four OCC AI risk tiers: How an AI use case is sorted from Tier 4 Low to Tier 1 Critical, with the OCC's own examples.
- The sixteen framework components: Each component in a paragraph: what it covers and the AI-specific risk the OCC attaches to it.
- Human oversight of AI and AI agents: Human-in-the-loop, human-on-the-loop and autonomous operation, and what makes oversight real.
- Board, management, audit and reporting: Who is accountable for a bank's AI, what the board sees, and what internal audit checks.
- Model risk management for generative AI: How the framework extends validation, monitoring, testing and documentation, and how it fits Bulletin 2026-13.
- Vendor and third-party AI: What examiners may look at when the AI belongs to a provider, including provider agents.
- Data, privacy and consumer compliance: Data governance, privacy, fair lending, BSA and what an AI system tells a customer.
- Resilience, kill switches and cybersecurity: Stopping, containing and recovering from an AI failure, and the security risks specific to agents.
- Preparing for an OCC exam of your AI: How examiners are told to use the framework and what a bank can have ready.
- What it means for community banks: The community-bank note, proportionality, and where a small bank's AI use usually falls.
- OCC AI glossary: The framework's lexicon, term by term, in plain English.
Is OCC Bulletin 2026-54 a regulation?
No. It transmits a framework written for OCC examiners. The bulletin says it does not establish supervisory expectations for banks and sets no enforceable standards or prescriptive requirements.
Does my bank have to follow the OCC AI framework?
The bulletin says banks are not expected to use the document to assess their AI use. Examiners may use it when they look at your bank's AI. Existing laws still apply to anything the AI does.
Is there a deadline to comply with the OCC AI framework?
No. The framework has no effective date and no compliance deadline. The OCC says it intends to update it as technology and supervisory understanding change.
Which OCC guidance covers machine-learning models?
Bulletin 2026-13, the revised model risk guidance of April 17, 2026. Footnote 1 of the framework says those principles apply to non-generative and non-agentic models. The new framework is more applicable to generative and agentic AI.
Does the OCC framework apply to community banks?
The bulletin's note for community banks says examiners may use it for community banks. They are told to tailor exams to the bank's size, complexity and risk profile, and to the nature of its AI use.
Does the OCC treat AI as a new risk category?
No. The framework says examiners should not treat it as a separate risk category. AI risk is managed inside existing categories such as operational, compliance and credit risk.
What does SI mean in OCC documents?
SI stands for Super Intelligence. The OCC uses it for the technologies previously called artificial intelligence, citing Executive Order 14434 of October 2, 2026.
- OCC Bulletin 2026-54, Super Intelligence: Supervisory Framework — OCC, Nov 4, 2026 (occ.gov) · our page on OCC Bulletin 2026-54
- Super Intelligence Supervisory Framework (PDF) — OCC, Nov 4, 2026 (occ.gov)
- News Release 2026-97, OCC Issues Supervisory Framework on Bank Use of Super Intelligence — OCC, Nov 4, 2026 (occ.gov)
- OCC Bulletin 2026-46, Third-Party Risk Management: Proposed Guidance and Request for Comment — OCC, Sep 11, 2026 (occ.gov)
- Semiannual Risk Perspective, Spring 2026 — OCC, May 7, 2026 (occ.treas.gov) · our page on OCC Semiannual Risk Perspective, Spring 2026
- OCC Bulletin 2026-13, Model Risk Management: Revised Guidance — OCC, Apr 17, 2026 (occ.gov) · our page on OCC Bulletin 2026-13
- Remarks by Acting Comptroller Rodney E. Hood, AI in Financial Services — OCC, Apr 29, 2025 (occ.treas.gov) · our page on Acting Comptroller Hood, 'AI in Financial Services' (Apr 2025)
- OCC Bulletin 2023-17, Third-Party Relationships: Interagency Guidance on Risk Management — OCC, Jun 6, 2023 (occ.gov) · our page on OCC Bulletin 2023-17
- OCC Bulletin 2021-19, Interagency Statement on Model Risk Management for BSA/AML Systems — OCC, Apr 9, 2021 (occ.gov) · our page on OCC Bulletin 2021-19
- Request for Information on Financial Institutions' Use of Artificial Intelligence (OCC Bulletin 2021-17) — Federal Register, Mar 31, 2021 (federalregister.gov) · our page on 2021 Interagency AI RFI (OCC Bulletin 2021-17)
- OCC Bulletin 2011-12, Sound Practices for Model Risk Management — OCC, Apr 4, 2011 (occ.gov) · our page on OCC Bulletin 2011-12
- OCC Bulletin 1997-24, Credit Scoring Models: Examination Guidance — OCC, May 20, 1997 (occ.gov) · our page on OCC Bulletin 1997-24
When this rule moves, the next morning's brief says so
when one of these regulators moves, the next morning's brief says so · six sourced stories · 7 am PT · free
plus every regulation, bank and agent page update, the morning after · leave any morning