Preparing for an OCC exam of your AI: what examiners may consider

Published Nov 21, 2026 · Last updated Nov 21, 2026

OCC Bulletin 2026-54 gives examiners a framework for reviewing a bank's AI, which the OCC calls Super Intelligence. It does not set expectations for banks, examiners are told to scale their work to how much is at stake, and AI is examined inside existing risk categories, not as a new one. A bank can prepare by keeping evidence for each of the framework's six pillars: purpose, risk, control, oversight, resilience and assurance.

TRACKED DOCUMENTS: OCC Bulletin 2026-54 · OCC Bulletin 2026-13

Who is the OCC AI framework written for?

It is written for OCC examiners, not for banks. Bulletin 2026-54 says the framework is “intended as a resource to assist examiners in their supervisory activities.”

The OCC calls AI “Super Intelligence (SI)” in this framework. This page writes “AI” and keeps “SI” only inside quotations.

The bulletin is addressed to the chief executive officers of all national banks, federal savings associations and federal branches and agencies, to OCC department and division heads, to all examining personnel and to other interested parties. It was issued on November 4, 2026, with News Release 2026-97. The news release describes it as a framework “for its bank examination workforce”.

The executive summary (section 1) lists the business activities in view: credit decisioning, payments, fraud detection, customer service, operations, risk management and market activities. Footnote 1 adds that the model risk guidance in Bulletin 2026-13 applies to non-generative and non-agentic models. So the framework is “more applicable to generative and agentic SI.” Generative AI writes text, images or code. Agentic AI can take actions on its own, such as sending a message or moving money.

The bulletin lists what examiners may do with it. They may use it to find AI-specific points that add to existing model risk management, governance, technology, compliance and operational-risk practices. They may also use it to tailor their work to the nature and potential consequences of a bank's AI use cases.

Does a bank have to follow the OCC AI framework?

No. The bulletin says the framework “does not establish supervisory expectations for banks”, and banks are not expected to use it to assess themselves.

The bulletin's summary makes four separate points:

  • It “does not establish supervisory expectations for banks.”
  • “Banks are not expected to use this document to assess their use of SI or their SI risk management.”
  • Nothing in it is intended “to establish new regulatory expectations or set forth enforceable standards or prescriptive requirements.”
  • The OCC intends to update it “as the technology, use cases, and supervisory understanding change and mature.”

A bank still has reasons to read it. It shows the considerations an examiner has been handed for thinking about AI. Put simply, it is the examiner's reading list, published. The news release also restates a standing point: whether activities are traditional or use new technology, “the OCC expects that bank activities are conducted in a safe and sound manner and consistent with applicable laws and regulations.” That expectation comes from existing law, not from the framework.

For the full list of what the framework is not, see the section-by-section reference guide.

How are OCC examiners told to scale their review of a bank's AI?

By how much is at stake. The bulletin says the OCC has instructed examiners to tailor their examination “based on the materiality of the SI use case.” Materiality means how much harm the use case could do if it went wrong.

The texts name these factors:

WhereWhat examiners are told to weigh
Bulletin summaryThe materiality of the AI use case.
Bulletin, note for community banksThe bank's size, complexity and risk profile, proportionate to the nature, consequences, autonomy and complexity of its AI use cases.
Section 1Examiners “should apply judgement” with the bank's size, complexity and risk profile in mind.
Section 2 introductionThe framework “should be considered in a manner proportionate to the nature, consequences, autonomy, and complexity of the bank's SI use cases.”
Section 1.1Risk classification may be based on potential consequences, transfer of agency and the degree of human oversight, “rather than the technical sophistication of the SI used.”

Section 1.1 gives its own example: “a simple rules-based agent authorized to initiate transactions could present greater risk than a sophisticated model limited to internal drafting assistance.” Autonomy here means how far the system acts without a person checking it first.

In practice, a bank that uses a chatbot only to draft internal notes and a bank whose agent sends payments are in different places. Section 3 sorts use cases into four illustrative tiers. Our page on the four OCC AI risk tiers explains them. Smaller banks can read what it means for community banks.

Is AI a new risk category in an OCC exam?

No. The bulletin, the news release and section 1 all say examiners “should not treat SI as a separate and distinct risk category.” Banks are also “not required to define a separate SI risk appetite.” A risk appetite is the amount and kind of risk a board agrees the bank will accept.

Section 1 explains the reasoning. AI use “can introduce, amplify or modify risk in existing risk categories, such as operational, compliance and credit.” Those risks, “including SI's contribution to them, should be identified, measured, monitored and controlled according to the bank's risk management framework.”

Section 1.1 adds that the framework may help identify transmission channels, meaning the routes by which an AI problem turns into, say, a credit or compliance problem. It does this “while not inferring any requirement to aggregate SI risk.”

For example, a customer-service chatbot that quotes a wrong fee is a consumer compliance issue with AI as the cause. Section 2.8 names hallucinations about “fees, rates, account status, or policies” as a source of legal or customer-harm exposure. A hallucination is fluent output that is false.

What can a bank have ready for each of the six pillars?

Section 1.2 groups the sixteen components into six pillars. For each one, below is what the framework describes, then evidence a bank could keep on hand. The evidence lists are our suggestions, not OCC text. Each is tied to the section it comes from.

1. Purpose (section 2.1)

The framework describes a documented purpose: what the system is designed to do, what it is not designed to do, who may use it, what data and systems it may touch and what actions it could take. Section 2.1 says activity outside the approved purpose “may be treated as a control failure even when the resulting output or action appears reasonable.”

  • Our suggestion: a purpose statement per use case, written in testable terms (2.1).
  • Our suggestion: an AI inventory listing use case, owner, provider, permissions, deployment status, monitoring and fail-over (2.1).

2. Risk (section 2.2)

The framework describes assessing each use case before deployment and after material changes. Factors include decision consequence, reversibility, customer and financial impact, autonomy, data sensitivity, explainability, provenance, permission scope and interdependencies.

  • Our suggestion: a dated risk classification for each use case with the reasoning for its tier (2.2, 3.1).
  • Our suggestion: a note on shared providers or data that link several use cases (2.2, “aggregate exposure”).

3. Control (sections 2.3, 2.5 to 2.8, 2.11)

This pillar covers human review, model risk management, data governance, privacy, compliance and cybersecurity. Section 2.3 describes review procedures defined so that approval is “more than a nominal click.” Section 2.11 names prompt injection, which is hidden or malicious instructions in an input that change what the system does.

  • Our suggestion: written review procedures and reviewer authority for each human checkpoint (2.3).
  • Our suggestion: data lineage records, meaning where the data came from and what was done to it (2.6).
  • Our suggestion: vendor terms on whether bank data can train the provider's models (2.7).
  • Our suggestion: the list of agent permissions and credentials, managed like privileged users (2.11).

4. Oversight (sections 2.3, 2.4, 2.12, 2.16)

The framework describes named accountable executives and use-case owners, approval authority at lifecycle stages, monitoring metrics and thresholds, and risk-based internal reporting.

  • Our suggestion: a list of accountable owners and who approves deployment, change, suspension and retirement (2.4).
  • Our suggestion: monitoring dashboards with thresholds, such as hallucination, override and complaint rates (2.12).
  • Our suggestion: board and management reports that reconcile to the inventory and incident records (2.16).

5. Resilience (sections 2.10, 2.11)

Section 2.10 says resilience covers both loss of service and incorrect operation. An agent “could remain available while taking inappropriate or harmful actions.” It lists tools such as circuit breakers, rate limits and kill switches. A kill switch is a way to stop the system at once.

  • Our suggestion: a tested containment and fallback plan for consequential use cases (2.10).
  • Our suggestion: a customer remediation procedure for when a misfire affects customers (2.10).

6. Assurance (sections 2.9, 2.13 to 2.15)

The framework describes testing across the lifecycle, documentation that lets others reconstruct material decisions, third-party oversight and internal audit whose challenge may depend on independence from both the owners and the validators.

  • Our suggestion: test results, including adversarial testing and hallucination measurement (2.13).
  • Our suggestion: logs of inputs, outputs, tool calls, approvals and version changes (2.14).
  • Our suggestion: vendor contracts covering notice of new AI features, incidents and liability (2.9).
  • Our suggestion: audit reports on AI and evidence that findings were closed (2.15).

For more on each area, see the sixteen framework components, human oversight, board, audit and reporting, vendor AI and resilience and cybersecurity.

What is the question behind each pillar?

Section 1.2 gives each pillar one principal supervisory consideration. The middle column restates it as a question in our own words, and the right-hand column is our suggestion. Neither is OCC text.

PillarThe consideration as a question (our wording of section 1.2)What to have ready (our suggestion)
PurposeWhat actions is the system or agent permitted to execute?Purpose statement and AI inventory (2.1)
RiskWhat are the consequences if its output or action is not fit for purpose?Risk classification with reasons (2.2, 3.1)
ControlWhich controls may prevent unauthorized, inappropriate or unsafe actions?Review procedures, data lineage, privacy terms, permission lists (2.3, 2.5 to 2.8, 2.11)
OversightWho monitors it and who is accountable for its use?Owner list, approval rights, monitoring metrics, board reports (2.3, 2.4, 2.12, 2.16)
ResilienceCan the bank detect, contain, recover from and remediate failures while critical processes continue?Tested containment, fallback and remediation plans (2.10, 2.11)
AssuranceIs there evidence that controls continue to operate effectively?Test results, logs, vendor contracts, audit reports (2.9, 2.13 to 2.15)

What does the OCC AI framework not tell a bank?

It gives no exam schedule, no deadline and no effect on ratings. None of the bulletin, the news release or the framework mentions any of these.

  • No exam schedule. The texts do not say when or how often examiners will review AI.
  • No effective date or deadline. The bulletin is dated November 4, 2026, and the texts name no compliance date.
  • No rating effect. The texts do not say how AI findings affect a bank's supervisory ratings.
  • No scoring weights. Section 3 calls its tiering “illustrative” and lists factors without weights.
  • No fixed question list. The components give “illustrative considerations” (section 1), and the document is “a living document”.

The OCC says it intends to update the framework, so the OCC page is the place to check for revisions. For model risk specifics, see model risk management for generative AI. For data and consumer rules, see data, privacy and consumer compliance.

Will OCC examiners use the AI framework at my bank?

The bulletin says examiners may use it, including for community banks. It does not say when or how often. Examiners are told to tailor their work to the materiality of each AI use case.

Should my bank score itself against the OCC AI framework?

The OCC says banks are not expected to use the document to assess their AI use or their AI risk management. Many banks may still read it to see which questions examiners have in front of them. Using it that way is a choice, not an OCC requirement.

Do we need a separate AI risk appetite before an exam?

No. The bulletin, news release and framework all say banks are not required to define a separate AI risk appetite. AI risk is handled inside existing categories such as operational, compliance and credit risk.

Is the OCC exam evidence list on this page official?

No. The descriptions of each pillar come from the framework, with section numbers. The evidence items are our suggestions, each tied to the section that inspired it.

Does the framework apply to our traditional credit scoring model?

Footnote 1 says the model risk guidance in Bulletin 2026-13 applies to non-generative, non-agentic models. The framework is described as more applicable to generative and agentic AI.

Is there a deadline to be ready for an OCC AI review?

No deadline or effective date appears in the bulletin, the news release or the framework. The OCC describes the framework as a living document it intends to update.

Does a simple AI tool get the same scrutiny as a complex one?

Not necessarily. Section 1.1 says risk classification may be based on consequences, transfer of agency and human oversight rather than technical sophistication. A simple agent that can send payments could carry more risk than an advanced drafting tool.

  1. OCC Bulletin 2026-54: Super Intelligence: Supervisory Framework — Office of the Comptroller of the Currency, Nov 4, 2026 (occ.gov) · our page on OCC Bulletin 2026-54
  2. Super Intelligence Supervisory Framework (PDF) — Office of the Comptroller of the Currency, Nov 4, 2026 (occ.gov)
  3. News Release 2026-97: OCC Issues Supervisory Framework on Bank Use of Super Intelligence — Office of the Comptroller of the Currency, Nov 4, 2026 (occ.gov)
  4. OCC Bulletin 2026-13: Model Risk Management: Revised Guidance — Office of the Comptroller of the Currency, Apr 17, 2026 (occ.gov) · our page on OCC Bulletin 2026-13

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