AI GOVERNANCE
AI governance: being able to answer for the decision, two years later.
Long after the tool was chosen, someone with authority asks who approved it, what it could decide, what went in, who watched, and what is on file.
Which question are you asking?
| Your question | Where it starts | You receive |
|---|---|---|
| Which AI is in use, deciding what, approved by whom? | AI Governance Assessment | Five named registers, the AI Policy among them, and a gap list |
| One vendor switched on an AI feature. Can we allow it? | AI Vendor and Third Party Review | A decision record for the vendor file |
| What may staff and vendors do with AI? | AI Policy and Acceptable Use | The policy, also in the Assessment |
| A model shapes a lending decision. Can we give the reason? | AI in Regulated Decisions, where fair lending sits | A specific reason for each decision |
| Who keeps answering for this every quarter? | AI Governance Program | Register, approvals, monitoring, board report |
- Smallest first step: one vendor.
- Usual first step: the Assessment.
- Continuing ownership: the Program.
Your charter and counterparties decide what already reaches AI.
Four of these reach AI because they reach every third party or every system holding the data, so they reached vendor AI the day it shipped; the fifth, the two GSE mandates of 2026, is about AI by name.
| Instrument | Why it matters here | Who it reaches, and who it does not |
|---|---|---|
| Interagency Guidance on Third-Party Relationships: Risk Management, 88 FR 37920 | A bank: its vendor program already reaches its AI. | Reaches Every bank the OCC, Federal Reserve and FDIC supervise, no asset size threshold. Status Guidance, not a rule. Not Credit unions; NCUA is not a party. |
| NCUA 12 CFR Part 748, Appendix A | A credit union, often handed the bank guidance by mistake. | Reaches Every federally insured credit union, no asset size exemption. Vendors Appendix A, III.D, with 07-CU-13 and Supervisory Letter 07-01. Not Banks. |
| NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers | An insurer: which state comes first. | Reaches Insurers in a state whose department adopted it, with that state's variations. Status Guidance, not statute. Not Insurers in a state that has not adopted it. Whether it reaches producers, managing general agents or third party administrators is a state question. |
| FTC Safeguards Rule, 16 CFR 314.4(f) | A mortgage lender or servicer. | Requires Service provider selection, contracts and periodic assessment, 314.4(f), at any consumer count; 314.6 does not remove them. Reaches Nonbank financial institutions under FTC jurisdiction. Not Banks, savings associations, federally insured credit unions. |
| Freddie Mac Guide Section 1302.8 and Fannie Mae LL-2026-04 | A Seller/Servicer: since 6 August 2026 both counterparties require it. | Requires Freddie Mac Guide 1302.8, from 3 March 2026, and Fannie Mae LL-2026-04, from 6 August 2026, require a Seller/Servicer using AI or machine learning in origination or servicing to hold a written governance framework for it. Reaches Any size, through the contract. Not Anyone outside either contract. |
One regulator says so in writing.
NYDFS industry letter on cybersecurity risks arising from artificial intelligence, 16 October 2024
Regulated entities are to address AI enabled social engineering, AI enhanced attacks, the exposure of the nonpublic information AI systems gather, and AI supply chain risk. The guidance imposes no new requirements beyond 23 NYCRR Part 500.
Holders of a DFS license, registration, charter or similar authorization, as Part 500 reaches them, including the 500.19 limited exemption, which the letter does not widen. Not an AI regulation.
The voluntary frameworks, and what a certificate does not say.
NIST AI Risk Management Framework (AI 100-1) and Generative AI Profile (AI 600-1)
NIST AI 100-1, released in January 2023, is voluntary: a filing structure, used where a client wants a recognizable shape. NIST states that version 1.0 is being revised under the 2025 White House AI Action Plan; no revised version had been published when this page was last reviewed, and the obligations that bind will not change because of it. Its Generative AI Profile, AI 600-1 of July 2024, is voluntary too.
Nobody by operation of law, and no regulator examines against either. A counterparty may adopt their structure by contract, and Freddie Mac's Section 1302.8 uses the NIST framework's terms of mapping, measuring and managing AI risk.
ISO/IEC 42001:2023
Certification means an accredited body assessed how an organization governs AI, not that any obligation was met.
Voluntary. No financial regulator in this market requires it or examines against it, and a certificate is a statement about a management system rather than about any AI system.
This firm builds the program and the evidence to the shape the instrument requires. Counsel confirms the legal reading for your state.
What is different when the governance work is done.
| Before | After |
|---|---|
| A tool list of whatever people volunteered | An inventory, each entry owned and tiered |
| Approval in an unwritten conversation | A recorded decision: what was weighed, by whom |
| A vendor's AI feature found in a release note | A vendor review before it is switched on |
| A supervisory question answered by reconstruction | An answer from a record kept as it happened |
Advice that is not tied to a product sale.
Our advice is not tied to product sales. If Red Stick Solutions ever holds a commercial relationship with a product or platform it discusses with you, that relationship is disclosed to you in writing before it can affect the advice.
Red Stick Solutions will not assess or test a security program it owns as virtual CISO for the same client in the same period. Where a client wants both, the assessment or the test goes to a third party and this firm helps select one.
So the answer can be a policy, a smaller project, or no purchase.
An AI system is also an attack surface.
A model gathers the data the security program protects into one place, in volume, within a vendor's reach. The design work is Cyber Defense. Neither practice is a component of the other; see Cybersecurity.
The first step starts with where AI already is.
Tell us what raised it: a board, an examiner, a vendor's new feature, or an unapproved use.