Risk Management

OCC Bulletin 2026-54, Explained: The OCC’s Super Intelligence (SI) Framework for Examining Bank AI

adapfin Team
adapfin Team
10 min read
OCC Bulletin 2026-54, Explained: The OCC’s Super Intelligence (SI) Framework for Examining Bank AI

On November 4, 2026, the Office of the Comptroller of the Currency (OCC) issued OCC Bulletin 2026-54, “Super Intelligence: Supervisory Framework.” It gives OCC examiners a common way to evaluate how banks use AI, which the OCC now calls Super Intelligence (SI). It covers generative AI and AI agents that take action on their own.

The framework is a resource for examiners. It doesn’t set new rules for banks. It does show what examiners are likely to look at, and what evidence they are likely to ask for.

Key takeaways

  • What it is: OCC Bulletin 2026-54 is a supervisory framework that helps OCC examiners evaluate banks’ use of SI (AI) in normal-course business activities.
  • What it isn’t: The OCC says the framework “does not establish supervisory expectations for banks.” It creates no new regulatory requirements.
  • The core idea: Control intensity could increase with the potential consequences of an AI agent’s failure or misuse.
  • Risk is about consequence: A simple rules-based agent that can initiate transactions could present more risk than a sophisticated model limited to internal drafting.
  • Structure: Six pillars (Purpose, Risk, Control, Oversight, Resilience and Assurance) group 16 framework components.
  • Model risk: The April 2026 model risk guidance that replaced SR 11-7 leaves generative and agentic AI out of scope. This framework is where the OCC addresses them.
  • Vendors: Accountability for managing the bank’s risk generally remains with the bank, even when a vendor builds or hosts the AI.
  • Community banks: Examiners may use the framework at community banks, tailored to the bank’s size, complexity and risk profile.

What is OCC Bulletin 2026-54?

OCC Bulletin 2026-54 announces a supervisory framework for evaluating banks’ use of SI. It is addressed to the chief executive officers of national banks, federal savings associations, and federal branches and agencies, and to all examining personnel. It is signed by James M. Gallagher, Senior Deputy Comptroller and Chief National Bank Examiner.

The framework itself is a separate document, linked from the bulletin as the Super Intelligence Supervisory Framework (PDF). It has four parts:

  • an executive summary with one overarching principle and six pillars,
  • sixteen framework components, each with key concepts and illustrative considerations,
  • an illustrative risk-tiering methodology with four risk tiers, and
  • a lexicon of SI terms.

The OCC calls the framework a living document. It expects to update it as the technology, use cases and supervisory understanding change. The framework also notes that it was prepared with the assistance of SI, and that a human reviewed, edited and approved the content (Framework §1).

What the framework is not

The bulletin is unusually direct about its limits. Three of its sentences matter most:

  • The framework “does not establish supervisory expectations for banks.”
  • It is “intended as a resource to assist examiners in their supervisory activities.”
  • Banks are not expected to use it to assess their own use of SI or their SI risk management.

The bulletin also says nothing in the guide is intended to establish new regulatory expectations, or to set enforceable standards or prescriptive requirements.

Two further points shape how examiners will apply it. Examiners should not treat SI as a separate and distinct risk category. Banks are not required to define a separate SI risk appetite. Instead, the OCC frames SI as something that can introduce, amplify or modify risk in existing categories, such as operational, compliance and credit risk (Framework §1).

Reading the framework still tells a bank a lot. It is the examiner’s lens. Knowing the questions in advance is a reasonable way to prepare for them.

What “Super Intelligence (SI)” means

“Super Intelligence” is the OCC’s term for what most of the industry still calls AI. The framework defines it in a footnote as the technologies and systems previously encompassed by the term “Artificial Intelligence” as defined in 15 USC 9401(3), along with the promise, potential and rapidly advancing capabilities of those technologies. The footnote cites Executive Order 14434 (October 2, 2026) (Framework §1, note 2).

In practice, read “SI” as “AI.” The lexicon uses parallel terms throughout (Framework Appendix A):

  • Generative SI is SI that generates synthetic content such as text, images, audio and video.
  • Agentic SI is SI capable of independently making decisions, interacting with its environment and optimizing processes without direct human intervention.
  • An SI agent is a system that autonomously perceives its environment, decides what to do and takes actions to achieve its goals.

How it relates to SR 11-7 and model risk management

The framework’s first footnote points to OCC Bulletin 2026-13, the revised interagency model risk management guidance issued April 17, 2026. That guidance replaced the 2011 model risk guidance, known as SR 11-7 at the Federal Reserve and OCC Bulletin 2011-12 at the OCC. It states that generative AI and agentic AI models are not within its scope.

The SI framework’s footnote says the model risk management principles in Bulletin 2026-13 apply to non-generative and non-agentic SI models. As a result, the framework is more applicable to generative and agentic SI (Framework §1, note 1). SR 11-7’s successor covers traditional models. This examiner framework is where the OCC addresses generative and agentic AI.

The framework doesn’t discard model risk management for AI. Its Model Risk Management Extensions component says certain MRM principles may be instructive. It adapts them for systems that are non-deterministic, change continuously or rely on third-party models (Framework §2.5). The bulletin doesn’t say it rescinds or replaces any existing guidance.

The overarching principle: consequence, not sophistication

The framework rests on one principle. A bank’s control intensity could increase with the potential consequences of an SI agent’s failure or misuse (Framework §1.1).

Higher-consequence use cases may warrant:

  • greater restrictions on autonomy,
  • stronger controls,
  • more direct human oversight,
  • more independent validation, and
  • more robust resilience measures.

Risk classification may be based on potential consequences, transfer of agency and the degree of human oversight. It is not based on the technical sophistication of the SI used.

The OCC gives a clear example. A simple rules-based agent authorized to initiate transactions could present greater risk than a sophisticated model limited to internal drafting assistance. What the AI is allowed to do matters more than how advanced it is.

The six pillars and 16 components

The framework groups its sixteen components into six pillars. Each pillar names one supervisory question. Two components appear under two pillars each: Human-in-the-Loop and Human-on-the-Loop Controls (Control and Oversight), and Cybersecurity (Control and Resilience) (Framework §1.2).

PillarWhat it asksComponents
PurposeWhat actions is the SI system or agent permitted to execute?Defined Purpose and Permitted Use
RiskWhat are the consequences if its output or action is not fit for purpose?Risk Classification and Criticality Assessment
ControlWhich controls may prevent unauthorized, inappropriate or unsafe actions?Human-in-the-Loop and Human-on-the-Loop Controls; Model Risk Management; Data Governance; Privacy; Compliance; Cybersecurity
OversightWho monitors the system, and who is accountable for its use?Governance & Accountability; Human-in-the-Loop and Human-on-the-Loop Controls; Performance Monitoring; Reporting & Transparency
ResilienceCan the bank detect, contain, recover from and remediate failures while critical processes continue?Operational Resilience; Cybersecurity
AssuranceIs there evidence that the controls continue to operate effectively?Testing; Documentation & Auditability; Third-Party & Vendor Risk; Internal Audit & Independent Challenge

The framework’s summary table names one illustrative SI-specific risk per component (Framework summary table). A few stand out for anyone running AI agents:

  • Defined Purpose and Permitted Use: scope drift or unauthorized action without an explicit, testable purpose boundary. The framework adds that activity outside the approved purpose may be treated as a control failure, even when the result looks reasonable (Framework §2.1).
  • Operational Resilience: a system that remains available while acting incorrectly, with no tested containment, continuity, recovery or remediation path.
  • Documentation & Auditability: an inability to reconstruct material decisions, changes, tool calls, approvals or system behavior.
  • Testing: testing that confirms expected behavior but doesn’t search for unknown, adversarial or hallucination-related failures.

How the framework classifies AI risk

The framework’s illustrative risk-tiering methodology scores every SI and agentic use case against the same factors, regardless of business line. The goal is that a comparable risk gets a comparable control, wherever it sits in the bank (Framework §3.1).

The seven illustrative factors are:

  1. Decision consequence: Does the system inform a human decision, make a decision, or take an action directly?
  2. Reversibility: Can an error be caught and undone before harm occurs?
  3. Customer and financial impact: How large is the potential harm if the system is wrong?
  4. Autonomy level: Is a human in the loop, on the loop, or absent from real-time checkpoints?
  5. Data sensitivity: Does it touch regulated, confidential or personally identifiable data?
  6. Explainability: Can its output be traced to a rationale that meets legal and disclosure obligations, such as adverse action notices?
  7. Provenance: Is it built in-house, fine-tuned from a third-party base model, or consumed as a vendor service?

The factors combine into an illustrative composite tier (Framework §3.2):

Illustrative tierProfileOCC’s illustrative use cases
Tier 1: CriticalDirect action, low or no reversibility, high financial or customer impact, autonomous or partially autonomousAutonomous payment agent; real-time credit decisioning; fraud-hold release; market access
Tier 2: HighDirect action or high-weight decision support, human-on-the-loop, moderate-to-high impactSI-assisted underwriting; agent-initiated customer communications; AML alert disposition; code assistance for internal code or data access changes
Tier 3: ModerateDecision support with mandatory human-in-the-loop, limited data sensitivity, reversibleInternal document summarization for credit memos; agent-assisted research
Tier 4: LowNo customer or financial data exposure, no direct action, fully human-controlledInternal productivity tools; code assistance; general employee Q&A

The OCC labels these tiers illustrative, not mandatory. One line in the methodology is worth remembering: a fully autonomous, customer-facing payment agent sits at Tier 1 “regardless of how well it performs in testing.” Good test results don’t lower the tier. Consequence and autonomy set it.

Transfer of agency and human oversight

Transfer of agency is one of the three bases for classifying risk, alongside potential consequences and the degree of human oversight. The framework doesn’t define the phrase in its lexicon. Read with the autonomy factor, it points to a practical question: how much of the decision, and the action, has the bank handed to the system?

The framework separates two kinds of human oversight (Framework §2.3):

  • Human-in-the-loop generally means review and approval before an output or action is used or executed.
  • Human-on-the-loop generally means active monitoring, with the authority and practical ability to intervene.

The level of involvement may be calibrated to the use case’s potential consequences, autonomy, reversibility and operating volume.

A named reviewer isn’t enough on their own. The framework’s illustrative risk for this component is “a named human reviewer without practical authority, competency, capacity, or a meaningful review standard.” Its considerations spell out what meaningful oversight looks like:

  • Competency and authority: The reviewer may need the expertise, information and authority to reject, modify, pause or halt the system’s output or action.
  • Meaningful review: Approval should be more than a nominal click. High volumes or bulk approval could reduce review quality and contribute to alert fatigue.
  • Span of control: A bank may evaluate how many agents, accounts or actions one person can credibly monitor, and how much time they have to intervene.
  • Override and escalation: Overrides, reasons and patterns may be logged and reviewed. A misfire, hallucination or control breach may justify temporarily increasing human involvement.
  • Independence: Where appropriate, the reviewer may be separate from the team that built or owns the system.

Vendors and third parties

The framework recognizes that banks frequently get SI capabilities from third parties rather than building them. Its position on accountability is plain. Although development or hosting could be distributed across vendors, accountability for managing the bank’s risk generally remains with the bank (Framework §2.4).

The Third-Party & Vendor Risk component supplements existing interagency third-party risk management guidance. Its illustrative considerations include (Framework §2.9):

  • Provider agent visibility: Banks may define the authority, identity, permitted tools, data use and escalation limits for a provider’s agents that interact with bank systems and data.
  • SI incidents: A bank may define an SI incident vocabulary and notification timelines for events such as hallucination, misfire, drift or manipulation.
  • Liability: A bank should consider how contracts assign responsibility for model-related errors or losses, including customer harm.
  • Change and exit: Existing vendors may add SI features or change models inside products the bank already uses. A bank may require notice of new SI functionality and may reassess the relationship or retest before or when changes are made.

If your AI arrives inside a vendor’s product, this component applies to you.

What it means for community banks

The framework applies to community banks, proportionately. The bulletin’s note for community banks says examiners may use the framework for supervising community banks’ SI use.

The OCC has instructed examiners to tailor examinations to the bank’s size, complexity and risk profile. That tailoring is proportionate to the nature, consequences, autonomy and complexity of the bank’s SI use cases. The framework adds that examiners will likely encounter various levels of SI sophistication and should apply judgment (Framework §1).

In practice, proportionality cuts both ways. A community bank whose AI use is limited to internal drafting and research sits in the lower illustrative tiers, where lighter controls may be appropriate. A community bank that lets an agent release fraud holds, send customer communications or initiate payments with little human involvement is closer to Tier 1 or Tier 2, the illustrative tiers the OCC gives for those uses, whatever its asset size. Size shapes how an exam is scoped. The use case shapes the tier.

What examiners are likely to ask for

The framework doesn’t list exam questions. Its six pillars map to six practical ones, though. Ask them of anyone putting AI in your core, and be ready to answer them yourself for the AI you already run, including the AI your vendors have added:

  1. Purpose: What exactly is the agent allowed to do, and where is that written down?
  2. Risk: If it gets something wrong, how bad is it and can it be undone?
  3. Control: What systems and data can it reach, and does customer data stay protected before it ever reaches a model?
  4. Oversight: Which actions need a human, and does that person see what the agent saw?
  5. Resilience: What are its limits, who can stop it, and how fast?
  6. Assurance: Can our internal audit reconstruct any decision without calling you, and will you tell us before your AI changes?

The documentation component is the one to start with. The framework says records may support reconstruction of material outputs or actions even when the model can’t produce a rule-based rationale. Inputs, outputs, retrieved sources, tool calls, approvals, overrides and actions are the evidence (Framework §2.14). If an examiner asks what your AI did last Tuesday, that record is the answer.

Get the Agent Evidence Worksheet. A working tool for COOs, CROs, CISOs and compliance leaders, organized around the framework’s six pillars. It includes:

  • A one-page evidence checklist for each of the six pillars
  • The framework’s illustrative risk tiers, with everyday banking examples
  • An agent “purpose card” template you can fill in for each use case
  • Contract questions for vendors who put AI inside your systems
  • A sample evidence record for one complete agent task

Download the worksheet →

Frequently asked questions

What is OCC Bulletin 2026-54?

OCC Bulletin 2026-54, “Super Intelligence: Supervisory Framework,” was issued by the Office of the Comptroller of the Currency on November 4, 2026. It gives OCC examiners a framework for evaluating banks’ use of Super Intelligence (SI), the OCC’s term for AI, in normal-course business activities. The framework has six pillars, sixteen components, an illustrative risk-tiering methodology and a lexicon.

Does OCC Bulletin 2026-54 create new requirements for banks?

No. The OCC says the framework does not establish supervisory expectations for banks and is intended as a resource to assist examiners. Banks are not expected to use it to assess their own SI use, and it does not set enforceable standards or prescriptive requirements.

What does “Super Intelligence (SI)” mean in the OCC framework?

Super Intelligence (SI) is the OCC’s term for the technologies and systems previously called Artificial Intelligence, as defined in 15 USC 9401(3), along with their rapidly advancing capabilities. In practice, SI means AI. The framework also uses the terms generative SI, agentic SI and SI agent.

What are the six pillars of the OCC’s SI supervisory framework?

The six pillars are Purpose, Risk, Control, Oversight, Resilience and Assurance. Purpose asks what the system may do. Risk asks what happens if it fails. Control covers preventive controls. Oversight covers monitoring and accountability. Resilience covers detecting, containing and recovering from failures. Assurance covers evidence that controls keep working.

How does the OCC framework classify AI risk?

By consequence, not technical sophistication. Risk classification may be based on potential consequences, transfer of agency and the degree of human oversight. The illustrative methodology scores decision consequence, reversibility, customer and financial impact, autonomy, data sensitivity, explainability and provenance, and places each use case in one of four illustrative tiers, from Tier 1 (Critical) to Tier 4 (Low).

What is the difference between human-in-the-loop and human-on-the-loop?

Human-in-the-loop generally means a person reviews and approves an output or action before it is used or executed. Human-on-the-loop generally means a person actively monitors the system and has the authority and practical ability to intervene. The framework says the level of human involvement may be calibrated to consequences, autonomy, reversibility and operating volume.

Does the framework apply to community banks?

Yes, proportionately. Examiners may use the framework for supervising community banks’ SI use. The OCC has instructed examiners to tailor examinations to the bank’s size, complexity and risk profile, proportionate to the nature, consequences, autonomy and complexity of the bank’s SI use cases.

Who is accountable when a vendor provides the AI?

The bank, generally. The framework says that although development or hosting could be distributed across vendors, accountability for managing the bank’s risk generally remains with the bank. Its third-party component covers provider agents, SI incident notification, contractual liability and notice when vendors add or change SI features.

How does the framework relate to SR 11-7 and model risk management?

OCC Bulletin 2026-13, the interagency model risk management guidance issued April 17, 2026, replaced the 2011 guidance known as SR 11-7 and states that generative AI and agentic AI models are not within its scope. The SI framework says those model risk principles apply to non-generative and non-agentic SI models, so the framework is more applicable to generative and agentic SI. The bulletin does not say it rescinds or replaces any guidance.

Sources

This article is for information only and isn’t legal or regulatory advice.

adapfin Team

adapfin Team

adapfin Technologies

Insights from the adapfin team.

See the platform behind the thinking.

See it with demo data, or join the founding-partner cohort.

Request a demo