OCC Bulletin 2026-54 · Agentic AIYour examiner is about to ask what your AI did last Tuesday.
On November 4, the OCC gave its examiners a framework for evaluating how banks use AI, including agents that take action on their own. Here’s how to have the answer ready.
What changed
Agentic AI just got its exam questions.
April 2026
The revised model risk guidance replaced SR 11-7, and it explicitly set generative and agentic AI aside.
November 2026
The OCC’s new supervisory framework covers them, under the name Super Intelligence (SI). Six pillars: Purpose, Risk, Control, Oversight, Resilience and Assurance.
The part that matters
Accountability generally stays with the bank, even when a vendor builds or hosts the AI.
The framework is a resource for examiners. It doesn’t set new rules for banks, but it shows what evidence examiners are likely to ask for.
One AI worker, one bank task
Follow a deposit-hold exception from start to finish.
A customer deposits a large check. It’s held. An AI worker reviews the exception and recommends a release. Here’s what an examiner should be able to see at every step.
- Pillar: Purpose & Risk
Step 1: Assigned purpose
The bank deploys the worker to one process, deposit-hold exceptions, with a tool belt, an authority limit and a named human supervisor. Nucleus only hands it work from that process.
- Pillar: Control
Step 2: Permitted tools
It can read the hold and the account history, verify the item and release holds within its authority. When it asks for anything else, such as wire history, Nucleus denies the request and logs it. Customer data is tokenized before any model sees it.
- Pillar: Oversight
Step 3: Approval checkpoint
Up to $5,000 the worker acts and its supervisor reviews afterwards. This hold is $18,400, so the worker can only recommend, and its supervisor approves first, seeing what the worker saw and why. The approver’s own authority is checked too.
- Pillar: Resilience
Step 4: Controlled execution
Nucleus re-checks the approval and limits at the moment of release. The supervisor can pause the worker at any time, and because AI runs apart from money movement, pausing it never stops the bank.
- Pillar: Assurance
Step 5: Audit evidence
Nucleus, not the AI, writes the record: every action by every actor, allowed or denied, in an append-only log internal audit can rebuild without calling a vendor.
Governance isn’t a layer you add to AI. It’s part of the operating system.
Before you sign
Six questions to ask anyone putting AI in your core.
- Purpose
Purpose: What exactly is the agent allowed to do, and where is that written down?
- Risk
Risk: If it gets something wrong, how bad is it and can it be undone?
- Control
Control: What systems and data can it reach, and does customer data stay protected before it ever reaches a model?
- Oversight
Oversight: Which actions need a human, and does that person see what the agent saw?
- Resilience
Resilience: What are its limits, who can stop it, and how fast?
- Assurance
Assurance: Can our internal audit reconstruct any decision without calling you, and will you tell us before your AI changes?
The Agent Evidence Worksheet
Have the evidence before the examiner asks for it.
A working tool for COOs, CROs, CISOs and compliance leaders. Use it to review the AI you already run, including the AI your vendors have added.
- 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
Get the worksheet
See it run
Watch one AI worker do the job, with the controls switched on.
Nucleus BankOS is a banking operating system built by bankers, for bankers. In 30 minutes we’ll walk through the deposit-hold agent end to end: purpose, permissions, approval, execution and the evidence it leaves behind. We’re also selecting a small group of design partners to shape what comes next.
The OCC framework is a supervisory resource for examiners and doesn’t establish requirements for banks. This page is for information only and isn’t legal or regulatory advice.
AI you can bank on.