The Deregulatory Convergence: How the CFPB, the DOJ, the FTC, and Colorado Simultaneously Dismantled AI Lending Oversight in 2026

The Deregulatory Convergence: How the CFPB, the DOJ, the FTC, and Colorado Simultaneously Dismantled AI Lending Oversight in 2026

Two federal agencies, one federal lawsuit, and one state legislature all moved in the same direction within months — leaving algorithmic lending with fewer guardrails than at any point since AI entered credit decisions.


The Simultaneous Rollback

For most of the past decade, the regulatory architecture governing AI in lending was built on two pillars: the federal disparate impact doctrine, which let regulators challenge lending practices that produced biased outcomes regardless of intent, and a growing body of state AI laws that required companies to test their algorithms for discrimination before deployment. In the first half of 2026, both pillars cracked simultaneously — and from different directions.

The Consumer Financial Protection Bureau finalized a rule eliminating disparate impact as an enforcement tool under the Equal Credit Opportunity Act, shifting its focus exclusively to intentional discrimination. Weeks later, Colorado repealed its landmark AI anti-discrimination statute and replaced it with a disclosure-only framework that drops every substantive anti-bias obligation. The Federal Trade Commission then asserted that even Colorado’s watered-down replacement may be preempted by federal law. And the Department of Justice, for the first time in history, intervened in a private lawsuit to strike down a state AI law on constitutional grounds.

None of these actions were coordinated through a single statute or executive order. But together they produced a convergent result: the regulatory floor for AI in lending is lower now than at any point since algorithmic credit scoring became widespread.


The CFPB Eliminates Disparate Impact Enforcement

In 2025, the CFPB made sweeping changes to its fair lending program, guided by a series of executive orders from the Trump administration. The bureau announced on its official fair lending page that it “is no longer using disparate impact in its supervision or enforcement of fair lending laws” and that it “closed all elements of open supervision exams and enforcement investigations that relied on disparate impact liability.” It also terminated CFPB orders that had relied on disparate impact, and ceased consulting with institutions on special purpose credit programs that consider race, national origin, or sex.[1]

In April 2026, the CFPB finalized a formal rule amending Regulation B of the Equal Credit Opportunity Act to codify these changes. The rule removes the long-standing requirement for creditors to avoid practices with unnecessary and arbitrary disparate impact on protected classes, and prohibits special purpose credit programs that rely on race, national origin, or sex. The National Fair Housing Alliance, a fair housing advocacy organization, noted that the rule was finalized “despite widespread public opposition, including objections raised by everyday people, civil rights and consumer protection organizations, industry players, State Attorneys General, and Members of Congress” — opposition the CFPB itself acknowledged constituted a majority of public comments.[2]

The core logic of disparate impact enforcement is straightforward: it catches discrimination that no one intended. A lending algorithm that uses proxy variables — zip codes that correlate with race, education levels that correlate with wealth — can produce systematically worse outcomes for protected classes without any human decision-maker harboring discriminatory intent. Intentional discrimination, by contrast, requires proof of motive, which is exceptionally difficult to establish for a machine learning model trained on millions of data points. By eliminating disparate impact as an enforcement theory, the CFPB effectively immunized algorithmic lending from the one legal tool designed to detect the kind of bias AI systems are most likely to produce: the unintentional kind.


Colorado’s AI Act: From Landmark to Disclosure-Only

While the CFPB was dismantling federal enforcement, Colorado was retreating from what had been the nation’s most ambitious state-level AI anti-discrimination law. In May 2024, Colorado enacted the Colorado Artificial Intelligence Act (SB 24-205), the first comprehensive state law governing the development and deployment of AI systems. The law focused on “high-risk” AI systems that were a “substantial factor” in reaching “consequential decisions” — a category that explicitly included lending and financial services.[3]

SB 24-205 imposed substantive duties on both developers and deployers of AI systems. Deployers were required to implement a written risk management policy, complete pre-deployment impact assessments, conduct annual reviews of system performance, and notify consumers before using AI to make consequential decisions. The law imposed a duty of care to avoid “algorithmic discrimination” — defined as differential treatment that disfavors individuals based on protected characteristics — and required companies to document the data governance measures used in developing their systems.[4]

The law faced sustained industry opposition from the start. Critics argued the requirements were unworkable, would stifle innovation, and placed disproportionate burdens on smaller companies. Amendment efforts during the 2025 regular legislative session stalled. Governor Jared Polis convened a special session in August 2025, but proposed bills failed there as well. The original February 1, 2026 effective date was pushed to June 30, 2026.[4]

On April 9, 2026, xAI — Elon Musk’s AI company — filed a lawsuit in U.S. District Court for the District of Colorado seeking to enjoin the law. Two weeks later, the Department of Justice moved to intervene, filing its own complaint alleging that SB 24-205 violated the Equal Protection Clause of the Fourteenth Amendment by “requiring AI companies to prevent unintentional disparate impact” and by exempting certain forms of discrimination designed to advance “diversity.” Assistant Attorney General Harmeet K. Dhillon stated: “Laws that require AI companies to infect their products with woke DEI ideology are illegal.” This was the first time the federal government intervened in a lawsuit challenging a state AI law.[5]

On May 14, 2026, Colorado Governor Polis signed SB 26-189, the Automated Decision-Making Technology Act (ADMTA), repealing SB 24-205 and replacing it with a substantially narrower framework. The new law drops every discrimination-specific obligation from the original statute:[3]

  • No duty of care — The ADMTA does not require deployers or developers to avoid algorithmic discrimination. Instead, it provides that existing state discrimination laws (such as the Colorado Anti-Discrimination Act) apply to regulated uses of automated decision-making technology.[4]
  • No impact assessments — SB 24-205 required pre-deployment impact assessments and annual reviews. SB 26-189 eliminates both.[4]
  • No risk management policy — The original law required deployers to adopt and maintain a written risk management policy. The replacement strips this requirement entirely.[4]
  • Disclosure-focused compliance — The new law requires deployers to give consumers notice before using automated decision-making technology for consequential decisions, disclose adverse outcome rights within 30 days, and provide a right to human review and reconsideration of decisions.[3]

The shift is significant in its underlying logic: rather than requiring companies to proactively prevent bias, the new law asks only that they tell consumers when AI was used and offer a human review after the fact. The regulatory model moves from prevention to disclosure — from requiring companies to find and fix discrimination before it harms people, to requiring them to inform people they may have been harmed.

For financial institutions specifically, lending and financial services remain a “covered domain” under the ADMTA, meaning the law still applies to AI used in credit decisions. However, the compliance burden is substantially lighter than under the original act. The law also provides practical exemptions for tools used for anti-money laundering compliance, sanctions screening, fraud prevention, and identity verification — and does not require disclosures that would be prohibited by federal law, including the Gramm-Leach-Bliley Act.[6]


The Federal Preemption Campaign

The rollback did not stop with Colorado’s legislative reversal. The federal government moved to preempt remaining state AI laws through a multi-pronged strategy combining executive orders, litigation, and regulatory assertions.

Executive Order 14365 and the AI Litigation Task Force

In December 2025, President Trump signed Executive Order 14365, “Ensuring a National Policy Framework for Artificial Intelligence,” which specifically called out Colorado’s law as an example of states “requiring American companies to embed ideological bias within their AI models.” The order directed the attorney general to establish an AI Litigation Task Force charged with identifying and challenging state AI laws — a directive that materialized months later when the DOJ intervened in the xAI lawsuit.[3]

The FTC’s “Suppression of Accuracy” Theory

On July 1, 2026, the FTC issued a Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems. The statement’s core theory is that AI companies represent their systems aim to produce the best possible output, and consumers reasonably expect accuracy. If a company steers outputs “contrary to consumers’ reasonable expectations” — including “attempted compliance with a State law, such as Colorado’s recently revised Artificial Intelligence Act” — it “may deceive consumers in violation of Section 5 of the FTC Act.”[7]

Critically, the FTC acknowledged that Colorado had revised its law, but stated that even “the new version poses many of the same concerns” and may be preempted. The agency cited the ADMTA’s liability provisions as applying “pressure” on AI developers to “alter” model outputs in ways that could themselves constitute deception under federal law.[4]

The FTC’s preemption theory works through implied conflict preemption. The FTC Act does not expressly preempt state law, but the FTC argues that state laws requiring companies to suppress AI accuracy without adequate disclosure conflict with Section 5’s core purpose of protecting consumers from deception. As the law firm Eversheds Sutherland explained, the FTC’s position creates a compliance dilemma: companies operating in Colorado may face a choice between complying with state law by modifying outputs (potentially triggering FTC liability) or prioritizing accuracy (potentially triggering state liability).[8]

The FTC’s framework does offer a theoretical safe harbor: companies can disclose that their AI systems prioritize objectives other than raw accuracy. But the agency sets an exceptionally high bar — disclosures must be “clear and conspicuous,” cannot be buried in terms of service, must be “prominent, persistent and placed where users will actually see it,” and must “affirmatively dispel the notion the system gives the best possible answer.” A “prominent misrepresentation is unlikely to be remedied by a less prominent, subsequent disclosure.”[7] The practical catch, as Eversheds noted, is that disclosing your AI deprioritizes accuracy may undermine the product’s entire value proposition — forcing companies to choose between commercial harm and enforcement risk.[8]


The Combined Effect on AI Lending

Understanding the convergence of these actions reveals how thoroughly the oversight landscape has shifted for AI in lending:

  • Federal anti-bias enforcement is restricted to intentional discrimination. The CFPB will no longer pursue disparate impact claims, meaning algorithmic lending systems that produce biased outcomes without provable intent face no federal enforcement under ECOA’s primary regulator.[1]
  • The most ambitious state AI law has been hollowed out. Colorado’s ADMTA still covers lending but eliminates the proactive anti-discrimination duties — impact assessments, risk management, duty of care — that would have required lenders to test their AI systems for bias before deployment.[4]
  • Remaining state protections face federal preemption pressure. The FTC’s proposed policy statement signals that even disclosure-based state AI laws may be challenged as federally preempted, creating legal uncertainty for any state seeking to regulate AI outputs.[7]
  • The DOJ has established a precedent of intervention. The federal government’s intervention in the xAI lawsuit signals willingness to challenge state AI laws in court — a tool that can be deployed against any state that passes substantive AI anti-discrimination legislation.[5]

For a lender using AI to make credit decisions, the practical consequence is that there is currently no federal regulator actively testing their algorithm for unintentional bias, no state requiring them to conduct impact assessments before deployment, and a federal agency actively arguing that state efforts to require output transparency may be unlawful. The enforcement gap is not a single policy change — it is the product of four separate actions by three federal agencies and one state legislature, all moving in the same deregulatory direction within a six-month window.


State Divergence and the Fragmenting Landscape

Not every state is following Colorado’s path. The rollback has produced a fragmented landscape where states are moving in opposite directions:

  • Connecticut expanded. The Connecticut Artificial Intelligence Responsibility and Transparency Act focuses on AI use in employment decisions, amending the state’s anti-discrimination statute to preclude automated decision-making as a defense to discrimination claims and making proactive anti-bias testing a mitigating factor for courts to consider.[4]
  • Utah narrowed. Utah amended its Artificial Intelligence Policy Act in 2025 to narrow disclosure requirements and add a safe harbor for companies following recognized compliance frameworks.[4]
  • Texas softened. Texas softened its original AI law by requiring both intent and an adverse outcome for discrimination claims, raising the bar for liability.[4]
  • Virginia vetoed. Virginia’s governor vetoed the state’s AI bill in March 2025, calling it “burdensome” to industry.[4]
  • California maintained. California adopted detailed rules governing automated decision-making technology under the California Consumer Privacy Act in 2025, including a consumer opt-out right that Colorado’s revised law does not provide.[4]

The logic of this divergence is that states are reading the federal pressure campaign differently. Some see the DOJ intervention and FTC preemption theory as a signal to retreat; others see the absence of a federal AI law as a reason to act. For lenders operating across state lines, the result is a compliance patchwork where the same AI lending system may face impact assessment requirements in California, disclosure-only rules in Colorado, and no state-level AI requirements at all in Virginia.


What Financial Institutions Should Watch

Despite the deregulatory trend, Colorado’s ADMTA still takes effect on January 1, 2027, pending rulemaking by the Colorado attorney general. Financial institutions using AI in lending should note several remaining obligations:

  • Consumer notice before AI use. Deployers must provide notice that automated decision-making technology is being used before it materially influences a consequential decision, including the consumer’s right to request additional information.[3]
  • Adverse outcome disclosures. When a consumer receives an adverse outcome from an AI-influenced decision, deployers must provide a plain-language description of the decision, the role played by the technology, and the consumer’s rights — within 30 days.[3]
  • Human review rights. Consumers may request meaningful human review and reconsideration of AI-influenced decisions. Deployers must designate an individual trained to conduct such reviews with authority to override the system’s decision.[3]
  • Record-keeping. Both developers and deployers must retain compliance records for a minimum of three years.[3]
  • Enforcement penalties. Violations are treated as deceptive trade practices under the Colorado Consumer Protection Act, carrying penalties of up to $20,000 per violation, enforced exclusively by the attorney general with a 60-day cure period (which does not apply to knowing or repeated violations).[3]

The FTC’s preemption theory adds a layer of uncertainty on top of these requirements. The agency’s position that state-law compliance is not a defense to federal deception claims means companies must carefully document any AI output modifications and be prepared to show they were made for disclosed, legitimate reasons rather than as undisclosed attempts to route around Colorado’s law.[4]


The Bigger Picture

The 2026 deregulatory convergence in AI lending reflects a deliberate policy choice: that protecting AI innovation and national competitiveness outweighs the preventive enforcement of anti-discrimination law in algorithmic credit decisions. The CFPB’s retreat from disparate impact, Colorado’s replacement of anti-discrimination duties with disclosure requirements, the DOJ’s constitutional challenge to state AI law, and the FTC’s preemption theory all embody the same underlying premise — that requiring AI companies to alter their outputs to prevent bias constitutes a form of compelled inaccuracy that the federal government should not permit.

Whether this premise is sound depends on a question the current regulatory framework has chosen not to ask: whether algorithmic lending systems produce biased outcomes, and if so, who is responsible for detecting and correcting them. The prior regulatory approach assumed that companies should proactively test for bias and that regulators should enforce against it even when unintentional. The current approach assumes that companies should produce accurate outputs as defined by the model’s own objectives, that disclosure is a sufficient consumer protection, and that existing anti-discrimination laws — enforced only against intentional discrimination — are adequate to catch harmful bias.

For the millions of consumers whose creditworthiness is now assessed by algorithms, the practical question is whether anyone is watching the system for bias. In 2026, the answer from the federal government and the state that once led the country on AI policy is: less than before, and possibly not at all.


References

[1] [PRIMARY] Consumer Financial Protection Bureau. “Guaranteeing Fair Banking for All Americans and Fighting Debanking.” May 28, 2026. https://www.consumerfinance.gov/fair-lending/

[2] [ADVOCACY] National Fair Housing Alliance. “CFPB Rule Change Opens the Door to Widespread Lending Discrimination During the Nation’s Fair and Affordable Housing Crisis.” April 2026. https://nationalfairhousing.org/cfpb-final-rule/

[3] [SECONDARY] Skadden, Arps, Slate, Meagher & Flom LLP. “Colorado Repeals and Replaces Its AI Act.” June 2026. https://www.skadden.com/insights/publications/2026/06/colorado-repeals-and-replaces-its-ai-act

[4] [SECONDARY] Thomas, Sheena, and Alyssa T. Calcerano. “Colorado Narrows Its AI Law, but the FTC Says That’s Not Enough: What Should Companies Make of These Changes?” Arnold & Porter. July 23, 2026. https://www.arnoldporter.com/en/perspectives/advisories/2026/07/colorado-narrows-its-ai-law-but-the-ftc-says-thats-not-enough

[5] [PRIMARY] U.S. Department of Justice, Office of Public Affairs. “Justice Department Intervenes in xAI lawsuit Challenging Colorado’s ‘Algorithmic Discrimination’ Law.” April 24, 2026. https://www.justice.gov/opa/pr/justice-department-intervenes-xai-lawsuit-challenging-colorados-algorithmic-discrimination

[6] [SECONDARY] Cooley LLP. “The New Colorado AI Act: What Financial Institutions Need to Know.” 2026. https://cdp.cooley.com/the-new-colorado-ai-act-what-financial-institutions-need-to-know/

[7] [PRIMARY] Federal Trade Commission. “Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems.” July 1, 2026. https://www.ftc.gov/system/files/ftc_gov/pdf/ai-policy-statement_0.pdf

[8] [SECONDARY] Eversheds Sutherland. “FTC’s Policy Statement on ‘Suppression of Accuracy’ in AI Systems Takes Aim at States’ Efforts to Regulate AI: What It Means for Businesses.” July 9, 2026. https://www.eversheds-sutherland.com/en/united-states/insights/ft-cs-policy-statement-on-suppression-of-accuracy-in-ai-systems-takes-aim-at-states-efforts-to


AI automated and generated content. Human reviewed.


Listen to the Bears Discuss This

Episode 14 of AI Policy for the Bears

The Weekly AI Policy Brief

AI policy moves fast. Get the week’s developments, new trackers, and podcast episodes in one short email — free, ad-free, no spam.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *