The OCC's AI framework (Bulletin 2026-54) describes the board and senior management naming accountable executives and use-case owners, receiving information on material AI use, and possibly setting an AI risk appetite, though the bulletin says banks are not required to define a separate one. Internal audit may give independent assurance over the whole AI risk management setup and report findings to the audit committee or board. The framework is a resource for examiners and does not set requirements for banks.
TRACKED DOCUMENTS: OCC Bulletin 2026-54
What does the OCC framework say about boards, managers and auditors?
Four of the framework's sixteen components describe who decides, who answers for an AI system, who checks it and who hears about problems. They are section 2.1 (purpose and permitted use), 2.4 (governance and accountability), 2.15 (internal audit and independent challenge) and 2.16 (reporting and transparency).
The framework came with OCC Bulletin 2026-54 on November 4, 2026. The OCC calls AI “Super Intelligence (SI)”, and this page simply writes AI. The framework is written for OCC examiners, the people who inspect banks. The bulletin says it “does not establish supervisory expectations for banks”. So everything below describes what examiners may look at, not rules a bank has to follow.
In section 1.2 the framework groups its components into six pillars. Governance and reporting sit in the “Oversight” pillar, which “establishes who monitors the SI system or agent and who is accountable for its use”. Internal audit sits in the “Assurance” pillar. For all sixteen components see the sixteen framework components, and for the whole document see the section-by-section reference.
Why does every AI system need a written purpose and limits?
Because without a written purpose nobody can tell when the system has gone too far. Section 2.1 says effective AI risk management “begins with a clearly defined and documented purpose”, which gives a basis for spotting scope drift, unauthorized use and control failures.
Scope drift means a tool slowly being used for things it was never approved for. Section 2.1 lists what a bank may write down before launch and after a material change:
- what the system is designed to do, and what it is not designed to do
- who or what may use it
- the data and systems it may reach, and the actions it could take
- when a person needs to review its work or switch it off
The framework adds that these limits may be “stated in terms that can be tested” rather than as a general description. It also says activity outside the approved purpose “may be treated as a control failure even when the resulting output or action appears reasonable.”
For example, a chatbot approved to answer staff questions about internal policy starts being pasted into customer emails. Even if the answers look right, that is a use nobody approved. Section 2.1 gives a similar case: a tool approved for internal research “may warrant additional review before it becomes customer-facing or decision-relevant.”
Section 2.1 also links the purpose to the bank's inventory, its list of AI systems. The inventory may record the use case, owner, provider, permissions, deployment status, monitoring and fail-over arrangements, and be updated when they change.
Who is accountable for an AI system or agent?
Named people inside the bank, even when a vendor built the system. Section 2.4 describes the board and senior management identifying “accountable executives and use-case owners”, and says accountability for managing the bank's risk “generally remains with the bank” when vendors develop or host the system.
Section 2.4 explains why this needs attention: “Rapid deployment, technical complexity, and distributed development or ownership could make accountability more difficult to identify.” The framework's summary table names the risk for this component as “Diffuse ownership or insufficient technical fluency limiting effective oversight and challenge”.
Section 2.4 also describes approval authority. Roles and decision rights may be defined for each step of a system's life: development or purchase, risk classification, testing, launch, material change, exceptions, suspension and retirement. Put simply, for each of those moments someone should be able to say who signs off. An AI agent is software that can take actions on its own, such as sending a message or moving data. The same accountability applies to it. Vendor questions are covered in vendor and third-party AI.
What does the framework describe for the board and senior management?
Setting the limits, naming the owners and receiving information on material AI use. Section 2.4 says the board and senior management “may establish an SI risk appetite, identify accountable executives and use-case owners, assign responsibilities and receive information on material SI use.”
It adds that governance may be supported by a complete inventory and by approval authority “at meaningful lifecycle gates”, meaning the key decision points in a system's life. Two further points concern directors directly:
- Knowledge. Board members, senior management, independent risk management, validation and audit “may need technical knowledge and access proportionate to the bank's use cases.” Validation is a check, based on evidence, that a model meets the requirements of its intended use.
- Reports that allow challenge. Board and management reporting may present material use cases, changes, exceptions, incidents, customer harm, concentrations and remediation “in a form that supports timely challenge and action.”
In practice, a board of a bank that uses AI only for internal drafting needs far less technical depth than one whose bank runs an AI agent that releases fraud holds. The framework ties the knowledge to the bank's own use cases, not to a fixed standard.
Does a bank need a separate AI risk appetite?
No. The bulletin's highlights say the framework “clarifies that examiners should not treat SI as a separate and distinct risk category, and banks are not required to define a separate SI risk appetite.” Section 1 of the framework says the same.
A risk appetite is the board's statement of how much risk the bank is willing to take. Section 1 explains that AI can “introduce, amplify or modify risk in existing risk categories, such as operational, compliance and credit.” Those risks are handled through the bank's existing risk management framework. Section 1.1 adds that the framework may help link AI risk management to risk appetite “while not inferring any requirement to aggregate SI risk.”
Section 2.4 still says the board and management “may” establish an AI risk appetite, and that it “could address autonomy, permitted actions, customer and financial impact, legal and operational exposure, and concentration in addition to model-performance measures.” Put simply, a bank may set one or may fold AI into its existing appetite. For example, an operational risk limit could include how many customer accounts an AI agent may change without a person's approval.
What do internal audit and independent challenge look at?
Whether the whole AI risk management setup works, not only single models. Section 2.15 says validation checks whether a system “operates as designed”, while internal audit “could provide additional assurance regarding governance, validation, control design, implementation, and the operation of the broader SI risk-management framework.”
Independent challenge means questioning by people who have no stake in the answer. Section 2.15 says effective challenge may depend on independence from both the team that builds or owns the system and the team that validates or monitors it. Its examples:
| Topic (section 2.15) | What the framework describes |
|---|---|
| Independence | Reporting lines, staffing, pay and prior involvement may be considered when judging whether audit can challenge objectively. |
| Access | Audit may need documentation, data lineage, validation results, vendor contracts, monitoring, incidents, overrides, remediation and underlying evidence “rather than only management summaries.” |
| Cadence and coverage | More consequential, fast-changing or highly autonomous use cases may warrant more frequent or deeper coverage. |
| Backstops | When a control agent, fallback agent, circuit breaker or human process is the safety net, audit may assess whether it is independently governed, tested and effective. |
| Reporting and closure | Findings may go to the audit committee or board, and closure may be supported by independent evidence that the problem was fixed. |
A circuit breaker is an automatic stop that halts a system when something looks wrong. For example, if a bank relies on a second AI tool to catch the first tool's mistakes, audit may ask who tests that second tool. More on stops and fallbacks is in resilience, kill switches and cybersecurity. How validation itself changes for generative AI is in model risk management for generative AI.
What AI reporting does the framework describe, and to whom?
Risk-based internal reports to the board and management, plus whatever notices laws, regulations or contracts call for. Section 2.16 says effective oversight depends on “timely and sufficiently detailed information” about material AI use cases, changes, incidents, failures and customer harm.
Section 2.16 lists what internal reporting may include:
- material use cases and their accountable owners
- significant changes and control exceptions
- incidents and customer remediation (putting things right for affected customers)
- concentrations, such as many systems relying on one provider
- assurance results, such as validation and audit findings
It says notifications required by laws, regulations and contracts “could be applicable to SI use cases.” It names three practices:
- Incident thresholds. The bank may set internal escalation and notification thresholds based on severity, duration, customer impact, financial exposure, legal requirements and potential systemic effects.
- Use-case and change reporting. Reviewing a bank's reporting may help supervisors understand consequential or highly autonomous use cases, material changes in purpose or permissions, and reliance on significant providers.
- Information quality. Reports may be reconciled to the AI inventory, incident records, validation results, audit findings and board reporting so that material differences are found.
In practice, the last point means the number of AI systems in the board pack should match the number in the inventory. The audience named in the text is the board and management (section 2.4), and the audit committee or board for audit findings (section 2.15). Section 2.16 does not set a reporting frequency.
Who does what in AI governance under the framework?
The table sums up roles only where the framework text names them. The framework does not use the term “business owner”. The closest term is “use-case owner” in section 2.4.
| Role | What the framework describes | Section |
|---|---|---|
| Board | May establish an AI risk appetite, identify accountable executives and owners, assign responsibilities, receive information on material AI use, may need technical knowledge proportionate to use cases, and may receive audit findings. | 2.4, 2.15 |
| Senior management | The same governance roles as the board in section 2.4, plus reporting that supports timely challenge and action. | 2.4 |
| Use-case owner | Named as accountable for the use case. The inventory may link each system to its owner, and reports may name accountable owners. | 2.1, 2.4, 2.16 |
| Independent risk management and validation | May need technical knowledge and access proportionate to the use cases. Validation assesses whether a system operates as designed. | 2.4, 2.15 |
| Internal audit | Assurance over governance, validation, controls and the whole framework, independent of builders, owners and validators, with access to underlying evidence. Findings may be reported to the audit committee or board. | 2.15 |
The framework does not assign specific tasks to a compliance function in these sections. Compliance topics are covered in data, privacy and consumer compliance.
What questions could a director ask management about AI?
These are our suggestions, not the OCC's. Each is built on a section of the framework so a director can see where it comes from.
- Which of our AI systems are material, and who is the named owner of each? (2.4, 2.16)
- Does each system have a written purpose that says what it may not do? (2.1)
- Has any system been used outside its approved purpose, and how did we find out? (2.1)
- Who can approve a launch, a material change, or a suspension? (2.4)
- How does our existing risk appetite cover AI, for example limits on what an agent may do alone? (2.4)
- What AI incidents and customer harm have we had, and what was fixed? (2.16)
- Does the AI count in this report match the inventory? (2.16)
- Does internal audit see the evidence itself, or only management summaries? (2.15)
- What training would help this board understand our AI use cases? (2.4)
What an examiner may ask is covered in preparing for an OCC exam of your AI. The OCC's AI documents are collected on the OCC AI regulation page.
Does the OCC require a board to approve every AI system?
No. The framework does not set requirements for banks. Section 2.4 describes approval authority at meaningful lifecycle gates and the board receiving information on material AI use, but it does not say the board approves each system.
Can a bank hand accountability for an AI system to its vendor?
Section 2.4 says that although development or hosting could be spread across vendors, accountability for managing the bank's risk generally remains with the bank. A vendor can build the system, but the bank still answers for how it is used.
Do board members need to be AI experts?
The framework does not say experts. Section 2.4 says board members and others may need technical knowledge and access proportionate to the bank's use cases. A bank with simple internal tools needs less depth than one running autonomous agents.
Is AI a new risk category under the OCC framework?
No. The bulletin and section 1 say examiners should not treat AI as a separate and distinct risk category. AI risk is handled inside existing categories such as operational, compliance and credit risk.
How is internal audit's role different from validation?
Section 2.15 says validation assesses whether an AI system operates as designed. Internal audit may give wider assurance over governance, validation, control design and the whole AI risk management framework.
How often should AI be reported to the board?
The framework gives no frequency. Section 2.16 says reporting may be risk based and timely, and section 2.4 describes reports in a form that supports timely challenge and action.
What counts as scope drift?
Using an AI system for something outside its approved purpose. Section 2.1 says such activity may be treated as a control failure even when the result looks reasonable.
- 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
- Super Intelligence: Supervisory Framework (full text, PDF) — Office of the Comptroller of the Currency, Nov 4, 2026 (occ.gov) · our page on OCC Bulletin 2026-54
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