By Roman Goldstein,, senior director at Klaros Group
Subscribe Now!
Stay on top of all the latest news and trends in the banking industry.
The Office of the Comptroller of the Currency has a long history of defending preemption of state laws that interfere with federally chartered banks’ ability to carry out their authorities.
A new Comptroller and a new Executive Order suggest that history may soon extend to national banks’ use of artificial intelligence.
Key insight: Preemption of state AI laws would give national banks a structural advantage over state banks, with the advantage compounding with implementation of every new state AI law.
National banks facing this Gordian Knot of compliance will soon profusely thank Abraham Lincoln and the Civil War-era Congress.
Key insight: The National Bank Act of 1864 has preempted state laws interfering with national bank operations for a century and a half. Its logic is simple: national banks operate under a federal charter, exercise federal authority, and thus cannot be balkanized by 50 different state regulatory regimes.
Courts have consistently upheld this framework, most prominently in Barnett Bank v. Nelson (1996). In that case the U.S. Supreme Court held that a state law is preempted when it “prevents or significantly interferes” with a national bank’s exercise of its powers.
That standard is the lens through which every state AI law affecting national banks will be evaluated.
Based on everything Comptroller Jonathan Gould has said and done since taking office, he will likely conclude that many state AI laws don’t apply to national banks.
Read more: Big Bank’s AI Layoffs Highlight the Human Cost of Poor Messaging
Some perspective here: Gould has made federal preemption a priority in a way no Comptroller has in many years.
In November 2025, he outlined a three-part preemption strategy:
• Proactively file friend of the court briefs defending the OCC’s preemption regulations.
• Work with the Treasury, OCC’s parent agency, to update those regulations.
• Mobilize industry and political allies to rebuild the political consensus underlying preemption.
“One of the key pillars of an economic union is federal preemption, which enables nationwide markets to arise,” Gould said months later.
That posture is reinforced by President Trump’s December 2025 executive order making it a national priority to create a federal framework for AI.
Among other things, the order directed the Department of Justice to establish an AI Litigation Task Force to challenge state AI laws that are inconsistent with a federal framework.
Key perspective: But litigation is slow — individual lawsuits, case-by-case outcomes, years of uncertainty. The OCC’s rulemaking authority is faster.
A Comptroller who has publicly committed to updating preemption regulations, working with Treasury, and filing aggressively in court won’t leave the state AI law questions unanswered.
Case study: Indeed, the OCC recently went to bat for preemption of the Illinois Interchange Fee Prohibition Act (IFPA), which forbids banks from charging interchange fees on taxes and tips paid by credit and debit card.
To make the case that the National Bank Act preempts IFPA, the OCC:
• Filed amicus briefs in federal district and appellate courts arguing IFPA was preempted.
• Sought public feedback on preempting IFPA, spurring industry and academic support for preempting the law.
• Made a formal preemption determination that IFPA didn’t apply to national banks.
• Amended the regulation governing national banks’ non-interest fees to strengthen the case for preemption.
Comptroller Gould did exactly what he promised to do last November.
Read more: How Autonomous AI Agents Will Really Redefine Banking Growth
Before the OCC can act on AI laws, it must answer a critical threshold question with significant procedural consequences: Are state AI laws “consumer financial laws” within the meaning of 12 U.S.C. § 25b?
• If yes, the OCC is constrained to case-by-case preemption determinations, subject to periodic five-year review, with state attorneys general retaining concurrent enforcement authority.
• If no, the OCC can make categorical determinations.
Section 25b defines “state consumer financial laws” as laws “that directly and specifically regulate the manner, content, or terms and conditions of any financial transaction … or any account related thereto, with respect to a consumer.”
Key insight: Many state AI laws don’t fit that description.
For example, the Texas Responsible Artificial Intelligence Governance Act is better understood as a general commercial regulation prohibiting certain AI architectures. California’s SB 53 targets frontier AI developers, not consumer-facing financial transactions. These operational laws are plausible candidates for categorical OCC preemption determinations outside the § 25b framework entirely.
Consumer-facing laws, such as Utah’s AI disclosure requirements, Colorado’s consumer appeal rights for high-risk AI decisions, and California’s forthcoming CCPA automated decision-making opt-out regulations, almost certainly fall inside § 25b and will require more deliberate case-by-case treatment.
Comptroller Gould knows this distinction well; his December 2025 escrow preemption proposal and IFPA preemption determination engaged carefully with the § 25b framework and its procedural requirements.
Looking ahead: Expect the OCC to sort incoming state AI laws along precisely this line, moving fastest on those it can address categorically.
Read more: Retail Bankers Are Adopting AI for All the Wrong Reasons
The preemption test, whether a state law “prevents or significantly interferes” with a national bank’s exercise of its powers, produces different answers for different categories of AI law.
• Machine-learning underwriting is the most straightforward target.
Lending is a core national bank power.
The OCC has already recognized that banks use alternative data and algorithmic credit models in its fair-lending guidance. State laws requiring human review of credit decisions, restricting permissible model inputs, or mandating explainability standards that machine learning models cannot satisfy by design directly interfere with a bank’s lending power.
This is where an OCC preemption determination would be easiest to defend and most useful to the industry.
• Employment AI regulations are a harder case, and the OCC will likely recognize that.
National banks already comply with a wide range of state employment laws, like paid leave mandates, wage withholding requirements, and anti-discrimination statutes. Illinois HB 3773 is structurally a civil rights law applied to a new technology. Courts have consistently declined to preempt that category of state regulation under the National Bank Act.
The OCC is unlikely to pick this fight — and if it did, it would probably lose.
• AI disclosure mandates present another steep climb.
A disclosure requirement doesn’t prohibit conduct; it layers an additional obligation on top. The First Circuit has set a demanding threshold, requiring either inevitable encroachment on federal banking power or an express conflict between federal and state law. A standalone disclosure mandate clears neither bar easily.
The more likely OCC play is to first establish a uniform federal AI disclosure standard. Once such a standard exists, state AI disclosure requirements that deviate from it face a tougher conflict preemption argument. One federal district court found California’s minimum payment disclosure mandate preempted because the OCC determined it impaired the efficiency of national banks.
Read more: The Identity Dilemma: How AI Blurs the Line Between Reality and Fraud
A related, and just as consequential, interpretive move the OCC could make involves a question courts have not yet addressed: Can AI agents that function as nondeterministic, autonomous software bind their employers?
If so, then national banks could operate through AI in the same way as they operate through human employees. The Uniform Electronic Transaction Act, enacted in every state but New York, leaves it to the courts whether an autonomous AI agent can be an electronic agent that can bind its employer, according to the Act’s commentary.
The OCC could (and under the Executive Order should) provide national uniformity by confirming that national banks can act through AI agents. That wouldn’t technically be a preemption determination (under § 25b or otherwise), but it would have preemptive effects, as explained below.
The National Bank Act authorizes a national bank to exercise its powers “by its board of directors or duly authorized officers or agents.” The statute lists agents alongside officers as the operative mechanisms of exercising power, without limiting either to natural persons. Indeed, the OCC treats operating subsidiaries — juridical persons — as instrumentalities through which national banks exercise their powers.
An AI agent authorized by a bank’s board or officers to originate loans, open accounts, execute transactions, or interact with customers on the bank’s behalf fits the ordinary meaning of “agent” at least as well as a corporate subsidiary does.
Furthermore, OCC regulations expressly allow a national bank to “perform, provide, or deliver through electronic means and facilities any activity, function, product, or service that it is otherwise authorized to perform, provide, or deliver.” An AI agent fits squarely.
If and when the OCC confirms national banks may employ AI agents, the preemption consequences will be significant: a bank’s deployment of an AI agent will constitute the exercise of a federal banking power; state laws restricting that deployment will face preemption.
Note that OCC regulations already say that “State laws that stand as an obstacle to the ability of national banks to exercise uniformly their Federally authorized powers through electronic means or facilities, are not applicable to national banks.”
Similarly, the OCC’s confirmation that an AI agent’s actions are attributable to the employing national bank would trigger the E-SIGN Act and its preemption provisions.
That law, 15 U.S.C. § 7001(h), says that contracts and other records created by an electronic agent (“a computer program or an electronic or other automated means used independently to initiate an action or respond to electronic records or performances in whole or in part without review or action by an individual at the time of the action or response”) are valid and enforceable “so long as the action of any such electronic agent is legally attributable to the person to be bound.”
A key gap: The Act doesn’t explain how to determine if an electronic agent’s actions are legally attributable to the employer. It’s telling that Congress didn’t reference the Uniform Electronic Transactions Act here, unlike in other parts of E-SIGN. That gap lets the OCC supply the necessary legal attribution.
Against that backdrop, the Comptroller is well-positioned to confirm that (i) AI agents can function as national banks’ agents under the National Bank Act and (ii) AI agents’ actions are legally attributable to the employing banks. Doing so would provide needed legal certainty to national banks’ use of AI agents.
Regulatory déjà vu: Does this playbook sounds familiar? That’s because this is exactly how the OCC handled the Illinois interchange law. First, it wrote a regulation confirming a national bank can do something; then, it concluded a state law frustrated that authority.
The federal court in the Illinois interchange litigation concluded that the OCC regulation made all the difference. The court originally held the interchange law applied to national banks but reversed course after the OCC clarified national banks’ authority to charge interchange fees.
Read more: Your Customers Will Blame You When Their Shopping Bots Go Rogue
These developments, taken together, could reshape the dual banking system.
National banks have long exported interest rates under 12 U.S.C. § 85, as interpreted by the Supreme Court in Marquette National Bank v. First of Omaha Service Corp. (1978).
State banks can also export interest rates, but states can opt out of this exportation. National banks are not subject to any such opt-out, providing a federal advantage that state banks cannot match.
Strategic insight: Fintechs involved in consumer lending have already started to shift to national bank partners because of states’ attempts to restrict interest rate exportation by state-chartered banks.
OCC preemption of state AI regulation would mean that national banks operate under a single federal AI regime rather than a state patchwork. Preemption of state AI laws would give national banks another structural advantage with real economic value, which would compound with every new state AI law that takes effect. A firm making charter decisions today is making a decades-long bet on regulatory architecture.
The national charter is becoming the more attractive option for technology-forward firms.
The Comptroller has spoken out in favor of preemption for state banks. Gould may well push for federal AI regulation for state banks. Congress, however, has twice considered and rejected preempting state AI laws.
The National Bank Act lets national banks operate as genuinely national institutions, not as 50 different banks using the same name. State AI laws create precisely the kind of regulatory fragmentation the National Bank Act was designed to prevent.
Comptroller Gould has called the national banking system “one of the crown jewels of American finance.” Expect this Comptroller to defend the system against state AI laws.
Read next: Brett King: Bankers Have Five Years to Rebuild for Agentic AI, Or Perish
Prior to joining Klaros Group, where he is senior director, Roman Goldstein served on the staff of the Federal Reserve Board of Governors. There, as lead innovation policy analyst, he reviewed banks' plans to engage in novel activities, reviewed foreign bank licensing applications, and developed supervisory policy for banking as a service, open banking and crypto. Before that, he served as an associate general counsel for Coinbase, where he advised the cryptocurrency company on compliance with banking and securities laws. In that capacity, he secured regulatory approvals worth $275 million in revenue per year, including first-of-their-kind product approvals. He also oversaw licensing for Coinbase Custody Trust Company.
Join over 2,500 of the brightest minds in banking at the industry’s premier annual event — three days loaded with the big ideas and latest innovations that are transforming banking today.
The Financial Brand is your premier destination for comprehensive insights in the financial services sector. With our in-depth articles, webinars, reports and research, we keep banking executives up-to-date with the latest trends, growth strategies, and technological advancements that are transforming the industry today.
Leave a Reply