← Back to all posts

What Bank and Credit Union Examiners Now Ask About AI

By Rami Akeela, Ph.D., Founder and CEO, Nera Systems Last updated

Bank and credit union examiners now ask about AI as a routine part of examinations. They want to know where AI is used, who governs it, which vendors supply it, what data those vendors receive, and whether the institution can shut a system off. No AI-specific rule covers generative AI tools such as ChatGPT, Claude or Gemini, so the existing data-security and third-party risk rules apply.

Updated October 30, 2026: added the October Reuters report on examiner questions about AI agents and shutdown, corrected the description of the CSBS framework's components, and refreshed the status of the planned federal request for information.

Sometime in the first half of 2026, AI stopped being a special topic in bank exams and became a routine one. According to Reuters reporting in June, examiners from the OCC and the Federal Reserve now raise it in standard examinations.

Most of the questions are what you would expect. One of them is harder than it sounds, and most institutions answer it with the wrong document: what data do your AI systems and AI vendors actually receive?

This piece covers what examiners are asking, what current guidance says and leaves out, how the picture differs for credit unions and state-chartered banks, and the questions worth having documented answers to before your next exam.

AI is now part of routine exams

On June 12, 2026, Reuters reported, citing people familiar with the matter, that OCC and Federal Reserve examiners had begun asking about AI in routine bank examinations (Reuters report via Quartz). The reported questions cover:

  • How AI systems are governed, and what technical limits constrain them
  • How client data is safeguarded and how data access is controlled
  • Third-party and vendor oversight, including exposure through subcontractors
  • Whether a bank has a way to disable an AI system, often described as a "kill switch"
  • What the bank would do if an AI system failed, including how it would move away from a compromised vendor

Examiners are reported to be focused on higher-risk uses: lending, know-your-customer checks and sanctions screening.

The questions are described as fact-finding, a way for supervisors to understand how banks use AI, and not a new rulebook. Neither the OCC nor the Fed commented on the report.

Fact-finding still has consequences. What an examiner learns in this round shapes what they expect in the next one, and banks report being unprepared for part of it. In Wolters Kluwer's US Banking AI Risk and Governance Index, a survey of 230 US banking professionals released in May 2026, 72% said their institution was least prepared for reporting or shutting down an AI incident. Respondents who pointed to the shutdown side most often cited missing model kill-switch protocols, at 34.35% (Wolters Kluwer).

October update: examiners are asking about AI agents

On October 13, 2026, Reuters reported that US bank supervisors have extended their questions to autonomous AI systems. Citing people familiar with the inquiries, the report says examiners are asking whether a bank can track and audit the decisions its AI agents make, what keeps an agent within its authority, who monitors and approves its actions, whether people stay involved in key decisions, and how the bank would shut such systems down. Some of the questions also reportedly cover Chinese-developed technology and the use of AI agents in cyber defense (ComplexDiscovery summary of the Reuters report).

The OCC, the Federal Reserve and the FDIC declined to comment, and the sources said regulators are still gathering information and have not begun prescribing practices. The reporting relies on anonymous sources, so it describes where examiner attention is heading, not a requirement.

The direction matters for data. An agent that can act on records is an agent that has received them, so every question about limits and shutdown also depends on knowing what the system was given in the first place. The same gap shows up in agent integrations generally: most MCP connections control who may request a record, not what the model receives.

The April guidance left generative AI out

On April 17, 2026, the Federal Reserve, OCC and FDIC issued updated model risk management guidance (OCC Bulletin 2026-13, published by the Fed as SR 26-2). It replaces SR 11-7, the framework banks have used for model risk since 2011 (OCC news release 2026-29).

The most important line for this topic sits in the scope statement. The agencies describe generative AI and agentic AI as "novel and rapidly evolving" and place them outside the guidance. Banks are told to govern them under their existing risk management and governance practices. The release also states that the guidance does not set enforceable standards or prescriptive requirements, and that it is expected to be most relevant to banking organizations with more than $30 billion in total assets.

Fed Vice Chair for Supervision Michelle Bowman described the same situation in remarks at an FSOC AI roundtable on April 27: "Today, banks are relying on existing risk-management frameworks to guide their use of AI." She said the Fed had amended its model risk guidance "to clarify that it does not apply to generative or agentic AI," and that the agencies should assess whether supervisory guidance is "fit for the future." She also left an open question on the table: "How should we consider third-party risk-management expectations for vendor-provided AI tools or partnerships?" (Remarks as published by the Federal Reserve Board, May 1, 2026).

The agencies said in April that they plan to issue a request for information on model risk management and banks' use of AI, including generative and agentic AI, "in the near future." As of October 24, no such request had appeared in the Federal Register.

So for the tools staff actually use, there is no AI-specific rulebook. There are examiners asking questions, and a set of older rules that still apply.

The rules that already apply to your AI vendor

No AI guidance does not mean no obligations. Two existing frameworks do most of the work, and both turn on data.

The first is the Interagency Guidelines Establishing Information Security Standards, which implement the Gramm-Leach-Bliley Act (GLBA) for banks (12 CFR Part 30, Appendix B). They require an information security program designed to protect against unauthorized access to or use of customer information. They define a service provider as any entity that maintains, processes, or otherwise is permitted access to customer information through the services it provides to the bank. And they require banks to exercise due diligence in selecting service providers, require appropriate safeguards by contract, and monitor them where the bank's risk assessment calls for it.

Read that definition next to how a typical AI tool works. When an employee puts a loan file or a customer spreadsheet into a model, the model provider processes that customer information. On the plain language of the guidelines, that provider looks like a service provider, with everything that follows from it.

The second is the Interagency Guidance on Risk Management of Third-Party Relationships, issued June 6, 2023 by the Fed, FDIC and OCC (Federal Register notice). It says a banking organization remains responsible for activities conducted through a third party, and it lists the third party's information security program among the things due diligence should review.

Neither framework mentions AI. Both describe exactly the question examiners are now asking.

Why a vendor contract doesn't answer the question

Most institutions answer the data question with a contract. The enterprise agreement says the provider will not train on your data, will delete it after a set period, and will protect it while it holds it. These are real protections, and a bank should have them.

They answer a different question. A no-training clause controls what the provider does with your data after it receives it. A retention limit controls how long the provider keeps it. Neither changes the fact that the provider receives readable customer data every time someone uses the tool on it. The contract governs the provider's behavior. It does not change what the model sees.

That distinction matters because of how the examiner question is framed. Questions about data access are questions about architecture: what leaves your environment, in what form, and who can read it when it arrives. A contract can promise that authorized access won't be misused. It can't make the access smaller than it is.

There is a second problem. Many AI capabilities now arrive embedded in tools a bank already licenses: the core provider's chatbot, the loan origination system's summarizer, the productivity suite's assistant. Each may sit on top of a foundation model from another company. That is the subcontractor exposure examiners are asking about, and it is easy to miss in a vendor inventory that only lists direct relationships. Consumer-goods companies hit the same wall from the other direction, where a data license, not a regulator, says no to AI.

Credit unions: the same questions, a narrower toolkit

NCUA's position is that AI needs no special rules. Its AI resource page says credit unions may use AI tools and technologies, that existing regulations are technology-neutral and apply to AI use, and that a credit union using an AI vendor must conduct appropriate due diligence, including understanding the vendor's safeguards and controls. Examiners review AI within the existing supervisory framework, including internal controls and third-party due diligence.

The difference is in what NCUA can see. In a May 2025 report, the Government Accountability Office found that NCUA lacks the authority to examine technology service providers, even as credit unions rely on them more for AI-driven services, and that its model risk guidance is "limited in scope and detail." GAO first recommended in 2015 that Congress give NCUA that authority, and the recommendation remains open (GAO-25-107197). In March 2026 NCUA launched an AI resources page, but GAO found it does not provide detailed model risk management guidance.

In practice, that puts more weight on the credit union itself. If the examiner cannot examine your AI vendor, your own documentation of what member data that vendor can read is most of what the examiner has to go on. We cover what that file needs in Can NCUA Examine Your AI Vendor?.

State examiners now have a playbook

On September 16, 2026, the Conference of State Bank Supervisors (CSBS) released an AI Supervisory Framework for state examiners. State regulators supervise about 79% of the 4,233 FDIC-insured banks and savings institutions, so for most community banks this is the closest thing to an AI exam manual that exists (White & Case summary).

The framework is voluntary and creates no legal requirements, and each state decides whether and how to use it. It includes a core examiner guide with scoping questions and a document request list, an examiner work program, a supplement for nonbank entities, and a worksheet for tiering AI use cases by risk.

Two parts speak directly to data. The risk-tiering worksheet treats data sensitivity as a factor, and its highest tier is defined by use cases involving sensitive personal data. And the questions examiners are directed to consider include whether the institution's AI comes from vendors or outside platforms, and whether it has looked for AI embedded in vendor products it already runs (Banking Dive).

Questions to have answers to before your next exam

Based on what examiners are reported to be asking, and on the frameworks they are working from, these are the questions we would want documented answers to:

  1. Where is AI in use today, including inside vendor products we already license?
  2. For each use, what customer or member data does the model receive, and in what form?
  3. Which companies can read that data when it arrives, including the foundation model behind a vendor's tool?
  4. Does our information security program treat each of those companies as a service provider, with due diligence, contract terms and monitoring to match?
  5. Can staff put customer data into an unapproved tool, and how would we know?
  6. Who can shut each AI system off, and how fast? For agents, can we audit what they did?
  7. If an AI vendor is compromised, how quickly can we cut it off, and what did it already hold?

Most institutions can answer the first question. The second and third are where answers usually turn into a description of the contract instead of a description of the data path. The seventh is the one the Wolters Kluwer survey suggests banks are least ready for.

Where this gets solved

The access question gets much smaller when the data is protected before it reaches an AI model, the approach usually called confidential AI. Nera is built that way. AI can work on an institution's most sensitive data while that data stays readable only on the institution's own side, and Nera cannot read it either. For an examiner, the answer to "what can the AI access" becomes a description of the architecture, not a promise in a contract. It also changes the shutdown answer: cutting off a vendor that never held readable data leaves nothing to claw back.

If you sit in model risk, compliance or the CISO seat at a bank or credit union, and the AI question in your last exam was answered with a vendor contract, we would like to hear what the examiner asked. You can see how Nera works for financial institutions or book 20 minutes with Rami.

Sources

  • Reuters, June 12, 2026, on AI in routine bank exams (via Quartz)
  • Reuters, October 13, 2026, on examiner questions about AI agents (ComplexDiscovery summary)
  • OCC News Release 2026-29 and OCC Bulletin 2026-13, April 17, 2026 (OCC)
  • Remarks by Vice Chair for Supervision Michelle Bowman at an FSOC AI roundtable, April 27, 2026 (text as published May 1, 2026)
  • Wolters Kluwer, US Banking AI Risk and Governance Index, May 2026 (Wolters Kluwer)
  • Interagency Guidelines Establishing Information Security Standards, 12 CFR Part 30, Appendix B (eCFR)
  • Interagency Guidance on Third-Party Relationships: Risk Management, June 2023 (Federal Register)
  • NCUA, Artificial Intelligence (AI) resource page, last modified April 28, 2026 (ncua.gov)
  • GAO-25-107197, Artificial Intelligence: Use and Oversight in Financial Services, May 2025 (GAO)
  • Conference of State Bank Supervisors, AI Supervisory Framework, September 16, 2026 (White & Case; Banking Dive)

Rami Akeela

Rami Akeela, Ph.D., is Founder and CEO of Nera Systems, a confidential AI infrastructure company. He holds PhDs in Electrical Engineering and Computer Engineering and has spent 20+ years in engineering and applied research. Nera Systems builds confidential AI infrastructure for organizations that cannot afford for their most sensitive data to be exposed to a model, a vendor, or each other.

Frequently asked questions

Do bank examiners ask about AI in routine exams?
Yes. Reuters reported in June 2026 that OCC and Federal Reserve examiners had begun asking about AI in routine examinations, covering governance, client data safeguards, vendor oversight, and the ability to shut a system off. The questions are described as fact-finding, not a new rulebook.
Does SR 26-2 apply to generative AI like ChatGPT?
No. The April 2026 model risk management guidance (SR 26-2 and OCC Bulletin 2026-13) describes generative and agentic AI as novel and rapidly evolving and places them outside its scope. Banks are told to govern them under existing risk management practices.
Is an AI model provider a service provider under GLBA?
Possibly. The Interagency Guidelines define a service provider as an entity that maintains, processes, or is permitted access to customer information through its services to the bank. A model provider that receives customer data from a bank's staff or systems appears to fit that language, which brings due diligence, contract and monitoring obligations.
Can credit unions use ChatGPT or other AI tools?
Yes. NCUA says credit unions may use AI tools and that existing regulations are technology-neutral. A credit union using an AI vendor must conduct appropriate due diligence, including understanding the vendor's safeguards and controls.
Does a no-training clause protect customer data in AI tools?
It limits what the provider does with the data after receiving it, which is a real protection. It does not change what the provider receives. If readable customer data reaches the provider, the data path still has to be disclosed, assessed and monitored.
What is the CSBS AI Supervisory Framework?
A voluntary set of tools released by the Conference of State Bank Supervisors on September 16, 2026, to help state examiners assess AI at the institutions they supervise. It includes an examiner guide, a work program, a nonbank supplement and a worksheet that tiers AI use cases by risk, with sensitive personal data in the highest tier.