Which AI rules US banks and insurers will answer to in 2027
Federal guidance got lighter and, as of November 2026, federal preemption is still a proposal, so the binding AI rules for 2027 come from Regulation B, two state laws, insurance bulletins and the EU AI Act.
Where the 2027 rules actually come from
US banks, credit unions and insurers will not answer to one federal AI law in 2027. As checked on 16 November 2026 we found no such statute, and the federal executive branch is pushing the other way: Executive Order 14365, signed on December 11, 2025, sets out a policy against state AI laws it regards as burdensome and asks for a federal framework that would preempt them [16]. What firms will be held to instead is a stack of older obligations applied to new tools: model risk guidance from the bank supervisors, the adverse action rules in Regulation B, state insurance regulators' bulletins, two state laws that take effect on January 1, 2027 [9] [15], and the EU AI Act for firms that lend or insure in Europe.
Federal guidance got lighter in 2025 and 2026; Colorado narrowed its law but kept a date; insurance regulators kept adding states; the EU delayed its high-risk deadline without dropping it.
Bank supervisors: model risk guidance rewritten, and AI set aside
On April 17, 2026 the Federal Reserve, the OCC and the FDIC replaced the 2011 model risk guidance that the industry knew as SR 11-7 and OCC 2011-12; the Fed's letter says the new SR 26-2 guidance supersedes and replaces SR 11-7 [1]. The OCC's bulletin rescinds OCC Bulletin 2011-12, the model risk booklet of the Comptroller's Handbook and the 1997 bulletin on credit scoring models [3].
Three features matter for AI programs. First, the guidance is expected to be most relevant to banking organizations with more than $30 billion in total assets [1]. Second, it says it sets no enforceable standards and that non-compliance with it will not result in supervisory criticism [2]. Third, a footnote puts generative and agentic AI models outside its scope because they are "novel and rapidly evolving", while stating that its principles do apply to traditional statistical models and to non-generative, non-agentic AI models [2]. The agencies said they plan a request for information on model risk management that will consider banks' use of AI, including generative and agentic AI [4].
The guidance was issued by the OCC, the Federal Reserve and the FDIC [3]. The NCUA, which supervises federal credit unions, is not among the issuers, so credit unions take their model expectations from their own supervisor and from the consumer laws below.
Moraph's view
Moraph's view: removing the threat of supervisory criticism does not remove the work. A credit, fraud or capital model that fails in production is still a loss and still a safety and soundness question. The larger gap is generative and agentic AI: as of 16 November 2026 there is no federal banking guidance that covers it directly, so banks are setting their own controls for those tools.
Consumer credit: the explanation duty stays, the guidance went
Regulation B still requires a creditor that takes adverse action to give a statement of specific reasons, and ยง 1002.9(b)(2) says the statement "must be specific and indicate the principal reason(s) for the adverse action" [7].
What changed is the CFPB's commentary around it. On May 12, 2025 the CFPB withdrew a batch of guidance documents that included Circular 2022-03, on adverse action notices for credit decisions based on complex algorithms, and Circular 2023-03, on the proper use of the sample forms [5]. The notice says the withdrawal "is not necessarily final" and that the Bureau will keep reviewing the documents [5]. Then on April 22, 2026 the CFPB issued a final rule that removes the "effects test" from Regulation B and states that ECOA does not recognize disparate-impact liability; the CFPB's Regulation B page shows the regulation as most recently amended on July 21, 2026 [6].
For broker-dealers and investment advisers, the SEC withdrew its 2023 proposal on conflicts of interest in the use of predictive data analytics on June 12, 2025, and said it does not intend to issue final rules on it [21].
Moraph's view
Moraph's view: federal disparate-impact exposure under ECOA has narrowed, but fair lending testing of models is unlikely to disappear from bank programs in 2027. State laws, the Fair Housing Act and a future change of federal leadership all keep the question alive, and a model that cannot produce specific adverse action reasons fails Regulation B on its own terms.
States: Colorado narrowed, California arrives, insurers keep adding bulletins
Colorado's 2024 AI Act, SB 24-205, was enacted in 2024, and a 2025 bill, SB25B-004, pushed its requirements to June 30, 2026 [10]; and SB26-189, approved on May 14, 2026, repealed and reenacted the framework, before that date, as rules on automated decision-making technology [8]. As checked on 16 November 2026, the act takes effect January 1, 2027 and applies to consequential decisions made on or after that date, with a few provisions effective on passage [9].
The new law reaches lending and insurance decisions among other consequential decisions [8]. A deployer whose covered technology materially influences a decision with an adverse outcome must, within thirty days, describe the decision and the technology's role, explain how to request more information about the tool and its inputs, and explain the consumer's rights [9]. A creditor whose notice under ECOA and Regulation B (and, where it applies, the Fair Credit Reporting Act) also satisfies these requirements complies with them and need not send a separate notice [9]. Developers must give deployers documentation from January 1, 2027, and the Attorney General must adopt rules clarifying the post-adverse outcome disclosures on or before the same date [9]. The Attorney General enforces the act, it creates no new private right of action, and before January 1, 2030 the Attorney General must give a 60-day notice and opportunity to cure before filing an action [8].
Two provisions matter for financial firms. Insurers and affiliates that are subject to Colorado's existing insurance statute on external data and predictive models, section 10-3-1104.9, are deemed in compliance in the practice of insurance [9]. And the act does not require disclosure of nonpublic personal information in a way that would violate the Gramm-Leach-Bliley Act [9]. That section, 6-1-1708, otherwise exempts HIPAA covered entities and FDA-regulated medical devices from the core duties and names no exemption for banks or credit unions [9]. Separately, the definition of a consequential decision leaves out activities relating to technologies used for anti-money laundering, sanctions compliance and fraud prevention [9].
California's privacy regulator finished its rules on automated decisionmaking technology when the Office of Administrative Law approved them on September 22, 2025 [14]. A "significant decision" includes one that results in the provision or denial of financial or lending services, defined to include credit, deposit accounts and funds transfers, and a business already using such technology for significant decisions must comply by January 1, 2027 [15].
For insurers the state picture is broader and older. The NAIC adopted its model bulletin on insurers' use of AI systems on December 4, 2023, and its implementation map, with status as of November 13, 2026, lists 30 adopting jurisdictions: 28 states, the District of Columbia and Puerto Rico [13]. Five of those adoptions are dated 2026, the most recent being North Dakota on October 5, 2026 [13]. The same map lists California, Colorado and New York separately as having their own insurance-specific rules [13]. New York's Insurance Circular Letter No. 7, issued July 11, 2024, expects board oversight, testing for unfair discrimination before an AI system goes into production, and oversight of vendors, and says an insurer may not rely solely on a vendor's claim of non-discrimination [11]. A separate NYDFS letter of October 16, 2024 on AI-related cyber risk says it imposes no new requirements beyond the Part 500 cybersecurity regulation [12].
Federal preemption: an order, a recommendation and a proposal
As checked on 16 November 2026, the federal preemption effort rests on an executive order, legislative recommendations and a proposed FTC policy statement [16] [17] [18]. Executive Order 14365 told the Attorney General to set up an AI Litigation Task Force within 30 days, told Commerce to publish an evaluation of state AI laws within 90 days, tied remaining BEAD non-deployment funds to that evaluation, and told the FTC to issue a policy statement within 90 days [16]. It names one state law, Colorado's ban on "algorithmic discrimination" [16].
The White House's legislative recommendations of March 2026 ask Congress to preempt state AI laws that impose undue burdens, but not state laws of general applicability, including laws to prevent fraud and protect consumers [17]. They also ask Congress not to create any new federal rulemaking body to regulate AI and to work through existing regulatory bodies with subject matter expertise [17]. The FTC's proposed policy statement, dated July 1, 2026, cites Colorado's revised AI act as an example of a state law that could lead a company to steer outputs in a way that may deceive consumers [18]; as checked on 16 November 2026 we found it only as a proposal.
Moraph's view
Moraph's view: an executive order cannot repeal a state statute, and the White House's own recommendation leaves general consumer protection law with the states. As of 16 November 2026 we found no enacted federal statute that preempts state AI laws, and we did not locate the Commerce evaluation on an official Commerce page. For 2027 planning, we read the Colorado and California dates as live unless a court enjoins them or Congress acts.
The EU AI Act for firms with European operations
Annex III of the AI Act lists as high-risk the AI systems used to evaluate the creditworthiness of natural persons or establish their credit score, other than fraud detection, and the AI systems used for risk assessment and pricing of natural persons in life and health insurance [20]. The Digital Omnibus on AI, Regulation (EU) 2026/1744, was published in the Official Journal on July 24, 2026 and entered into force on the third day after its publication [19]. It moved the high-risk obligations for Annex III systems to December 2, 2027 and for Annex I product-related systems to August 2, 2028, and it added new prohibitions that apply from December 2, 2026 [19].
The 2027 stack at a glance
| Rule | Who it reaches | Status as checked on November 16, 2026 |
|---|---|---|
| Fed SR 26-2 and OCC Bulletin 2026-13, model risk management [1] [3] | Fed, OCC and FDIC supervised banks; most relevant above $30 billion in assets [1] | Issued April 17, 2026; non-enforceable; generative and agentic AI out of scope [2] |
| Regulation B adverse action notices | Creditors, including banks, credit unions and fintech lenders | In force; effects test removed by the April 22, 2026 rule [6] |
| Colorado SB26-189 [8] | Developers and deployers of technology that materially influences consequential decisions | Takes effect January 1, 2027 [9] |
| California CCPA rules on automated decisionmaking | Businesses covered by the CCPA using such technology for significant decisions | Compliance by January 1, 2027 [15] |
| NAIC model bulletin on AI systems | Insurers in adopting jurisdictions | Adopted in 30 jurisdictions [13] |
| NYDFS Insurance Circular Letter No. 7 | Insurers authorized in New York | Issued July 11, 2024 [11] |
| EU AI Act, Annex III high-risk | Credit scoring and life and health insurance pricing in the EU | Applies from December 2, 2027 [19] |
What to watch in 2027
Four events will decide how heavy 2027 feels. The Colorado Attorney General's rules on post-adverse outcome disclosures are due on or before January 1, 2027, and they may include sector-specific guidance and address how the disclosures interact with other federal and state notice laws [9]. As checked on 16 November 2026, the Department of Law filed proposed rules on August 11, 2026 and held a public rulemaking hearing on October 26, 2026, and its page did not yet show final rules [22]. The bank agencies' request for information on model risk and AI would be the first step toward federal expectations for generative and agentic AI [4]. Any final FTC policy statement, or a lawsuit by the federal task force against a state AI law, would test the preemption theory in practice [16]. And in Europe, the December 2, 2027 Annex III date sets the timetable for credit scoring and insurance pricing models used there [19].
Moraph's view
Moraph's view: the practical risk in 2027 is inconsistency, not a single heavy rule. A lender operating nationally can face specific-reasons notices under Regulation B, a Colorado post-adverse outcome disclosure that the Regulation B notice covers only if it also meets Colorado's content requirements, pre-use notices and opt-out handling in California, and EU high-risk duties for European portfolios, each with its own definition of the system in scope. A single inventory of models and decision tools, tagged by decision type and jurisdiction, absorbs these changes better than one program per law.
What this means for you
- Banks: Map models to the April 2026 guidance, which covers non-generative AI, and decide how generative and agentic tools are governed before the agencies' request for information arrives [2] [4]. Colorado's act lists a financial or lending service as a covered domain, and its section on other legal obligations, 6-1-1708, names no exemption for banks, so lending decisions made on or after January 1, 2027 that a covered technology materially influences are within its scope; a Regulation B notice that also meets the act's disclosure requirements counts as compliance [9].
- Credit unions: The interagency model risk guidance was issued by the OCC, the Fed and the FDIC, not the NCUA [3]. Regulation B's specific-reasons duty applies to credit union lending as to any creditor [7], and Colorado's consequential decision scope includes lending [8].
- Insurers: Check which of the 30 NAIC-bulletin jurisdictions and the separate California, Colorado and New York regimes apply to the book of business [13]. In Colorado, insurers subject to section 10-3-1104.9 are deemed in compliance with the new act in the practice of insurance [9].
- Fintech partners: Developers of decision tools sold into Colorado must give deployers documentation from January 1, 2027, and deployers' post-adverse outcome disclosures depend on that information [9]. NYDFS expects insurers not to rely solely on a vendor's claim of non-discrimination [11].
Moraph predictions
Every prediction is Moraph's view, not a fact. Each has a date, a confidence level (70% means we expect to be wrong about three times in ten) and a test anyone can apply. We score them on this page when the date comes and never remove one. How we make and score predictions.
P1 By Confidence 75% Open
The Colorado Attorney General adopts final rules implementing the post-adverse outcome disclosure requirements of SB26-189 on or before March 31, 2027.
How we will check: The Colorado Department of Law or the Colorado Secretary of State's Code of Colorado Regulations shows the ADMT rules adopted, with an adoption date on or before March 31, 2027.
Basis: SB26-189 directs the Attorney General to adopt these rules on or before January 1, 2027, and the act applies to decisions made from that date [9]. The Department of Law filed proposed rules on August 11, 2026 and held the rulemaking hearing on October 26, 2026 [22].
P2 By Confidence 70% Open
No federal statute that expressly preempts state laws regulating AI in consumer credit or insurance decisions is enacted by December 31, 2027.
How we will check: Search Congress.gov public laws enacted in 2027 for a provision expressly preempting state AI laws; the prediction is right if none covers credit or insurance decisions.
Basis: Executive Order 14365 and the March 2026 White House recommendations ask Congress for preemption but would leave general consumer protection laws with the states [16] [17].
P3 By Confidence 65% Open
The OCC, the Federal Reserve and the FDIC publish a request for information on model risk management that addresses banks' use of AI, including generative or agentic AI, by June 30, 2027.
How we will check: A Federal Register notice or joint release from the three agencies, dated on or before June 30, 2027, requesting information on model risk management and AI.
Basis: The agencies said in April 2026 that they plan such a request [4], and the revised guidance left generative and agentic AI out of scope [2].
P4 By Confidence 65% Open
The NAIC implementation map lists at least 33 jurisdictions as having adopted the AI model bulletin by December 31, 2027.
How we will check: Count the 'Adopted States' entries on the NAIC map 'Implementation of NAIC Model Bulletin: Use of Artificial Intelligence Systems by Insurers' in its latest version on or before December 31, 2027.
Basis: The map lists 30 adopting jurisdictions as of November 13, 2026, five of them adopted in 2026 [13].
P5 By Confidence 75% Open
The EU AI Act's high-risk obligations for Annex III systems, including credit scoring and life and health insurance pricing, apply on December 2, 2027 without a further postponement.
How we will check: On December 2, 2027, the consolidated AI Act on EUR-Lex shows December 2, 2027 as the application date for Annex III high-risk obligations, with no later amending regulation moving it.
Basis: Regulation (EU) 2026/1744 set the Annex III date at December 2, 2027 [19], and Annex III covers creditworthiness and insurance pricing systems [20].
Facts on this page link to their sources. Text marked "Moraph's view", and every prediction, is our own analysis, not a fact or advice. Methodology.
Spotted an error? Tell us and we will check it and correct the piece where needed.