America's first comprehensive state AI law was repealed before it ever took effect — undone by a coordinated legal campaign from Elon Musk's xAI, the Trump administration's Justice Department, and a federal court injunction. The replacement statute signed into law on May 14, 2026, carries lighter obligations but faces an uncertain future: both challengers are expected to test it in court as well.

A Two-Year Saga Ends in Retreat

Colorado's SB 24-205 — known as the Colorado AI Act — was signed in May 2024, making Colorado the first U.S. state to impose broad governance requirements on AI systems used in high-stakes decisions: hiring, housing, healthcare, education, insurance, and lending. It required developers and deployers to maintain risk management programs, conduct annual algorithmic impact assessments, and exercise a formal "duty of reasonable care" to prevent algorithmic discrimination.

Governor Jared Polis signed the bill "with reservations" and immediately invited the legislature to revise it. Two years and three legislative sessions of failed amendment attempts followed. A special session in August 2025 only managed to push the effective date from February 1, 2026, to June 30, 2026, buying more time.

The clock then ran out on legislative compromise — and a new legal strategy emerged.

The Federal Assault on SB 24-205

On April 9, 2026, xAI — Elon Musk's AI company, developer of the Grok language model — filed suit in the U.S. District Court for the District of Colorado (Case No. 1:26-cv-01515), arguing the law violated the First Amendment's prohibition on compelled speech, the Commerce Clause, and the Equal Protection Clause. xAI's complaint specifically targeted the law's carve-out exempting AI systems designed to "increase diversity or redress historical discrimination" from the definition of prohibited algorithmic discrimination.

The Trump administration moved with unusual speed. On April 24, 2026, the U.S. Department of Justice filed a complaint in intervention — the first time the federal government had ever sought to invalidate a state AI law, according to Norton Rose Fulbright's analysis, confirmed by the DOJ's own press release. The DOJ invoked the Civil Rights Act of 1964 (42 U.S.C. § 2000h-2) and alleged that SB 24-205 violated the Equal Protection Clause by compelling AI companies to treat users differently based on protected characteristics and by exempting diversity-advancing algorithms from the same anti-discrimination rules. "Laws that require AI companies to infect their products with woke DEI ideology are illegal," said Assistant Attorney General Harmeet K. Dhillon in the DOJ's press statement.

The intervention was a direct product of President Trump's December 2025 Executive Order 14365, which established a DOJ AI Litigation Task Force and explicitly named Colorado's SB 205 as a target, according to Jenner & Block's analysis of the order's provisions. Axios reported that Colorado's law was "the only state AI law specifically called out" in that executive order.

The same day the DOJ intervened, xAI and the Colorado Attorney General filed a joint motion to suspend enforcement. On April 27, 2026, Magistrate Judge Cyrus Y. Chung of the U.S. District Court for the District of Colorado granted the motion, freezing enforcement pending completion of any rulemaking and a ruling on xAI's forthcoming preliminary injunction request.

The Replacement: SB 26-189

Facing a June 30 effective date for a law already enjoined and constitutionally challenged, Colorado's legislature moved with unusual speed. SB 26-189 was introduced May 1, 2026, cleared the Colorado Senate 34-1 and the House 57-6, and was signed by Governor Polis on May 14, 2026. The law takes effect January 1, 2027, contingent on the Attorney General completing mandatory implementing rulemaking by that date.

The most consequential change is what the new law removes. SB 26-189 eliminates three of SB 24-205's most contested provisions, according to analyses from Seyfarth Shaw and Crowell & Moring:

  • The duty of reasonable care to prevent algorithmic discrimination — the core substantive anti-discrimination obligation — is gone entirely.
  • Mandatory risk management programs are eliminated.
  • Annual algorithmic impact assessments are no longer required.

In their place, SB 26-189 establishes a notice-and-transparency framework centered on "Covered Automated Decision-Making Technology" (ADMT) — technology that processes personal data to generate predictions, recommendations, classifications, or scores used to "materially influence" consequential decisions. The covered domains remain the same (employment, housing, healthcare, insurance, education, lending, and government services), but the obligations are lighter:

  • Pre-use notice: Deployers must inform consumers before ADMT is used in a consequential decision.
  • Post-adverse-outcome notice: If ADMT contributes to an adverse outcome, the deployer must provide the consumer a plain-language explanation within 30 days.
  • Consumer rights: Consumers may request correction of inaccurate personal data and, where commercially reasonable, meaningful human review of the decision.
  • Developer documentation: Developers must provide deployers with documentation covering intended uses, known harmful uses, training data categories, and system limitations, and must notify deployers of material updates.

The law also introduces a fault-based liability framework that allocates discrimination exposure between developers and deployers based on whether the ADMT was used as intended and whether the developer provided adequate documentation — directly addressing the liability ambiguity raised in Mobley v. Workday, according to Epstein Becker Green. Crucially, any contract clause purporting to shift a party's own discrimination liability to another party is voided under the new law.

A 60-day pre-enforcement cure period applies through January 1, 2030, after which the Attorney General can proceed directly to enforcement.

What Was Lost — and What Was Kept

The terminological shift from "high-risk artificial intelligence system" to "Covered ADMT" is substantively significant. ADMT requires no inference capability — a system that merely scores or ranks individuals qualifies, making the scope nominally broader. But by removing the governance architecture (risk programs, impact assessments, duty of care), the compliance burden is substantially reduced.

Consumer advocates who had supported the original law expressed concern. As the Colorado Sun reported, the two-year fight ended with a "watered-down law" and "little fanfare." The bill passed with bipartisan support, reflecting broad agreement that the original law needed revision even as critics disagreed about the direction.

SB 26-189 also explicitly closes a gap the original law left open: general-purpose LLMs deployed with acceptable-use policies (such as ChatGPT or Claude) are excluded from the ADMT definition when not used specifically for consequential decisions, according to Epstein Becker Green's analysis.

The Fight May Not Be Over

Despite the replacement, legal uncertainty persists. McDermott Will & Emery noted that "both xAI and the DOJ are expected to challenge SB 189 as well." The constitutional arguments — particularly the First Amendment compelled-speech theory and the Equal Protection challenge to any remaining diversity carve-outs — could apply to the new law's disclosure requirements.

The Colorado Attorney General has stated he will not enforce SB 26-189 until rulemaking is complete, and the federal court's April 27 order technically conditions xAI's preliminary injunction motion on the completion of that rulemaking. That means the litigation posture is unlikely to resolve before 2027 at the earliest.

For AI developers and deployers in other states, the implication is clear: the Trump DOJ's AI Litigation Task Force remains active, and the Colorado campaign represents the first successful test of a federal strategy to displace state AI governance through litigation. Other state laws — particularly those containing algorithmic anti-discrimination standards or diversity carve-outs — may be next on the Task Force's list. The Commerce Department was directed under EO 14365 to publish an evaluation of "onerous" state AI laws by March 11, 2026; that list has not been publicly released, according to Jenner & Block.