The OCC Tells Examiners Not to Treat Super Intelligence as a Separate Risk Category

On November 4, 2026, the Office of the Comptroller of the Currency issued a framework for its bank examiners. It covers how banks use what the regulator calls Super Intelligence, or SI. A footnote says the term refers to the technologies and systems "previously encompassed by the term" Artificial Intelligence, as federal law defines it. The term also takes in the "promise, potential, and rapidly advancing capabilities" of those technologies.

Another footnote says the framework itself "was prepared with the assistance of SI." The content, it says, "has been reviewed, edited, and approved by a human."

The framework says it should be treated as a living document. It expects to change as the technology, its uses and supervisors' understanding change. Examiners should apply judgment, with a bank's size, complexity and risk profile in mind. Its sixteen components, it says, describe how SI may affect established control areas. They identify illustrative considerations for applying risk-based supervisory judgment.

Parts of the framework follow, not the whole.

Not a Separate and Distinct Risk Category

The release says examiners "should not treat SI as a separate and distinct risk category." Banks are not required to define a separate risk appetite for SI. Elsewhere, the framework says a board and senior management may set one.

The framework says SI use can introduce, amplify or modify risk in existing categories, such as operational, compliance and credit. Those risks, including SI's part in them, should be identified, measured, monitored and controlled under the bank's own risk management framework.

A Reviewer Who Can Say No

Human oversight is most effective, the framework says, when the bank defines the nature of the involvement and can demonstrate that the control operates as intended. It separates two kinds. Human-in-the-loop generally means review and approval before an output or action is used. Human-on-the-loop generally means active monitoring, with the authority and practical ability to step in.

The assigned reviewer may need enough expertise, information and authority to evaluate, reject, modify, pause or halt the system's output or action. Review triggers, evidence and procedures may be defined so that approval is "more than a nominal click." High volumes or bulk approval, the framework says, could reduce review quality and contribute to alert fatigue. A bank may evaluate how many agents, accounts, processes, alerts or actions one person can credibly monitor, and the time available to step in.

For anyone who owns part of a bank or lends to one, that is the question to put: can the bank show that its reviewers had the authority, the skill and the time to say no?

When a System Keeps Running and Gets It Wrong

Operational resilience for SI, the framework says, includes both loss of service and incorrect operation. An agentic system "could remain available while taking inappropriate or harmful actions." That would require the bank to detect it, contain it, recover, remedy the harm and keep the underlying business process going.

For testing, the framework's summary table names an illustrative risk. It is testing that confirms expected behavior "but does not search for unknown, adversarial, or hallucination-related failure modes." The framework defines a hallucination as output that is erroneous or flawed but still in the form of a convincing narrative or presentation. Hallucinations about fees, rates, account status or policies, it says, could contribute to legal or customer-harm exposure.

Sized by Consequence

Risk classification may be based on potential consequences, transfer of agency and the degree of human oversight, the framework says, rather than on the technical sophistication of the SI used. A simple rules-based agent allowed to start transactions could carry greater risk than a sophisticated model limited to internal drafting assistance.

Its illustrative tiering method gives an example. A simple internal productivity tool with no customer or financial data exposure sits at Tier 4, which its table calls Low. A fully autonomous, customer-facing payment agent sits at Tier 1, which it calls Critical, "regardless of how well it performs in testing."