Alston & Bird Consumer Finance Blog

SI

OCC Issues Super Intelligence Supervisory Framework: What Banks, Nonbank Partners, and Secondary Market Participants Need to Know

On November 4, 2026, the Office of the Comptroller of the Currency (OCC) issued OCC Bulletin 2026-54, “Super Intelligence: Supervisory Framework,” along with a framework designed to support OCC examiners in evaluating banks’ use of “super intelligence” (SI) in normal-course business activities. The OCC uses SI to refer to the technologies previously encompassed by the term “artificial intelligence” as defined in 15 U.S.C. 9401(3), citing Executive Order 14434. The framework applies to activities including credit decisioning, payments, fraud detection, customer service, operations, risk management, and market activities.

What Happened?

The framework presents sixteen components grouped into six pillars: Purpose, Risk, Control, Oversight, Resilience, and Assurance. It also includes an illustrative risk-tiering methodology and a technical lexicon drawn largely from the AI Lexicon workstream of the Artificial Intelligence Executive Oversight Group. The OCC explains that model risk management principles in OCC Bulletin 2026-13 apply to non-generative and non-agentic SI models, so the new framework “is more applicable to generative and agentic SI.” The OCC also discloses that the framework was prepared with the assistance of SI and was reviewed, edited, and approved by a human.

The OCC is explicit about the framework’s status. The bulletin states that the framework “does not establish supervisory expectations for banks,” that banks “are not expected to use this document to assess their use of SI,” and that nothing in it is intended to “set forth enforceable standards or prescriptive requirements.” Examiners “should not treat SI as a separate and distinct risk category, and banks are not required to define a separate SI risk appetite.” The OCC calls the framework a living document and intends to update it as technology, use cases, and supervisory understanding mature.

Which Institutions Are Covered?

The framework applies to national banks, federal savings associations, and federal branches and agencies of foreign banking organizations. Examiners may apply the framework to community banks, but are instructed to tailor examinations to the bank’s size, complexity, and risk profile, proportionate to the nature, consequences, autonomy, and complexity of its SI use cases.

Nonbank lenders, servicers, and fintech partners are not directly covered. The framework states that accountability for managing the bank’s risk generally remains with the bank even when development or hosting is distributed across vendors. Nonbanks that provide SI-enabled services to, originate for, or service loans for national banks should therefore expect these considerations to arrive through their bank partners’ third-party oversight, due diligence, and contract negotiations.

What Does the Framework Cover?

The framework’s overarching principle is that a bank’s control intensity could increase in correlation with the potential consequences of an SI agent’s failure or misuse. Higher-consequence uses 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 rather than technical sophistication. As the OCC puts it, a simple rules-based agent authorized to initiate transactions could present greater risk than a sophisticated model limited to internal drafting assistance. Key components include:

  • Defined purpose and permitted use. Before deployment and after a material change, a bank may document what a system is and is not designed to do, who may invoke it, the data and systems it may access, the actions it could take, and when human review or suspension is required. Activity outside the approved purpose may be treated as a control failure even when the result appears reasonable.
  • Human-in-the-loop and human-on-the-loop controls. Human-in-the-loop means review and approval before an output or action is used; human-on-the-loop means active monitoring with authority and practical ability to intervene. Reviewers need competency and authority, review must be more than “a nominal click,” and banks may evaluate how many agents or alerts one person can credibly monitor. A misfire or hallucination may justify temporarily increasing human involvement.
  • Governance and accountability. The board and senior management may establish an SI risk appetite, identify accountable executives and use-case owners, and set approval authority across development, deployment, material change, suspension, and retirement. Board members, risk management, validation, and audit may need technical fluency proportionate to the bank’s use cases.
  • Model risk management extensions. Some SI does not fit traditional assumptions of model stability or reproducible outputs. The framework points to more frequent monitoring for drift and vendor updates, repeated testing and consistency thresholds for non-deterministic outputs, and alternatives to champion-challenger testing for generative outputs. It cautions that explainability tools “do not reveal the SI’s reasoning.”
  • Data governance and privacy. Banks may assess the lineage, licensing, and representativeness of training and production data, including data embedded in third-party foundation models, and subject synthetic data to quality and bias review. Privacy risks include model memorization, inference of sensitive characteristics, retention of prompts, outputs, and embeddings, vendor use of bank data to train models for other customers, and cross-border processing.
  • SI “does not change the bank’s underlying legal obligations.” The framework highlights Bank Secrecy Act and sanctions screening (where poor data or thresholds can increase false negatives or positives), consumer compliance in servicing, collections, disclosures, and marketing, and hallucinations about fees, rates, account status, or policies that could create legal or customer-harm exposure.
  • Third-party and vendor risk. Banks may define the authority, identity, permitted tool use, and escalation limits of provider agents; adopt an SI incident vocabulary with notification timelines covering hallucination, misfire, drift, or manipulation; consider how contracts allocate liability for model-related errors and customer harm; and require notice when vendors add SI features or change models.
  • Operational resilience. An agentic system “could remain available while taking inappropriate or harmful actions.” The framework cites monitoring, circuit breakers, rate limits, least-privilege permissions, and kill switches; fail-over to trained staff or narrower agents; a separate customer remediation workstream; and controls for external customer, fintech, or agentic-commerce agents interacting through bank channels.
  • Attack surfaces include prompt injection and jailbreaking, data and model poisoning, model extraction and data leakage, and excessive agency through broad credentials. Agent identities may be governed in a manner comparable to privileged users.
  • Testing, documentation, and audit. Testing methods include adversarial testing, red teaming, expert subjective review, LLM-as-a-judge, and hallucination measurement. Documentation should support reconstruction of material outputs through inputs, retrieved sources, tool calls, approvals, overrides, and version history. Internal audit coverage may scale to autonomy and rate of change, with access to underlying evidence rather than only management summaries, and findings reported to the audit committee or board.

How Does the Illustrative Risk Tiering Work?

The framework describes a graduated control model: “light-touch for low-risk, low-autonomy use, materially heavier for anything approaching customer-impacting, irreversible, or highly autonomous action.” Each use case may be scored on seven factors, regardless of business line: decision consequence, reversibility, customer and financial impact, autonomy level, data sensitivity, explainability (including adverse action notice obligations), and provenance (in-house, fine-tuned third-party, or fully external). The factors combine into four illustrative composite tiers:

  • Tier 1: Critical. Direct action, low or no reversibility, high financial or customer impact, autonomous or partially autonomous. Examples: autonomous payment agents, real-time credit decisioning, fraud-hold release, market access.
  • Tier 2: High. Direct action or high-weight decision support with human-on-the-loop oversight. Examples: SI-assisted underwriting, agent-initiated customer communications, AML alert disposition, code assistance for internal code or data access changes.
  • Tier 3: Moderate. Decision support with mandatory human-in-the-loop review, limited data sensitivity, reversible. Examples: internal summarization for credit memos, agent-assisted research.
  • Tier 4: Low. No customer or financial data exposure, no direct action, fully human-controlled. Examples: internal productivity tools, code assistance, general employee Q&A.

A fully autonomous, customer-facing payment agent sits at Tier 1 “regardless of how well it performs in testing.”

Why Does It Matter?

Although the framework creates no new requirements, it shows what OCC examiners will ask about generative and agentic SI, the area to which the OCC says the framework is most applicable. For banks, the framework provides a detailed benchmark for purpose statements, human oversight, containment, and evidence. For nonbank lenders and servicers, bank partners are likely to carry these expectations into diligence questionnaires and contracts, particularly on provider agents, incident definitions, model change notice, and liability. For structured finance and secondary market participants, SI used in underwriting, servicing, and collections is becoming a diligence item, and the ability to reconstruct SI-supported loan-level decisions may bear on representations, warranties, and repurchase exposure.

What Do I Need to Do?

Banks and their partners should use the framework to inform or enhance existing governance, model risk, compliance, and third-party programs account for SI. This should include the following steps:

  • Document purpose and permissions. Record each use case’s permitted actions, data and system access, decision boundaries, and escalation triggers in terms that can be tested.
  • Tier use cases by consequence. Score use cases on the seven factors and confirm that control intensity, validation, and audit coverage track the resulting tier.
  • Make human oversight real. Define reviewer authority and expertise, review standards, span of control, and override logging, and watch for alert fatigue and bulk approvals.
  • Build containment and remediation. For Tier 1 and Tier 2 uses, test kill switches, circuit breakers, fail-over processes, and customer remediation workflows.
  • Revisit vendor contracts. Address provider agent permissions, SI incident definitions and timelines, model change notice, restrictions on training with bank data, liability allocation, and exit rights.
  • Address compliance and privacy. Test for hallucinations in customer-facing representations, recalibrate BSA and sanctions tools, and set retention and access controls for prompts, outputs, and logs.
  • Expand testing and evidence. Adopt adversarial, red-team, and hallucination testing, and retain the inputs, tool calls, approvals, and version history needed to reconstruct material decisions.
  • Prepare for partner and investor diligence. Nonbanks and secondary market participants should expect, and be ready to provide or request, inventory, tiering, testing, and incident evidence.

Alston & Bird’s team is available to assist banks, nonbank lenders and servicers, and structured finance participants in assessing their SI governance programs against the OCC framework and in preparing for examinations and partner diligence.