xAI Sues Colorado While Colorado Quietly Prepares to Replace Its Own AI Act

Colorado's AI Act — the first comprehensive U.S. state law targeting high-risk AI — is simultaneously being attacked in federal court and rewritten in Denver.

On April 9, 2026, xAI filed a federal lawsuit asking a federal judge to declare the Colorado AI Act unconstitutional and permanently enjoin its enforcement before it takes effect on June 30. Three weeks earlier, on March 17, 2026, Governor Jared Polis's own AI Policy Workgroup unanimously endorsed a proposed rewrite that would, if enacted, scrap much of the current law and replace it with a narrower framework focused on automated decision-making technology (ADMT).

These two stories are less contradictory than they look. Both reflect the same pressure: the Colorado AI Act, as currently written, has more critics than defenders — including the governor who signed it. What is unclear is whether the law survives to its effective date in its current form, gets replaced by the ADMT Framework, gets blocked by a federal court, or some combination of all three.

Here is what each development actually does, what it does not do, and what deployers should be doing right now.

The Lawsuit: X.AI LLC v. Weiser

xAI filed its complaint in the U.S. District Court for the District of Colorado on April 9, 2026. The case is *X.AI LLC v. Weiser*, No. 1:26-cv-01515, naming Colorado Attorney General Philip J. Weiser in his official capacity as the defendant responsible for enforcement.

xAI is asking the court for two things: a declaratory judgment that Senate Bill 24-205 is unconstitutional, and a permanent injunction blocking the Attorney General from enforcing it. A preliminary injunction, if granted before June 30, would freeze enforcement while the litigation proceeds.

xAI's Four Legal Theories

The complaint raises four constitutional arguments. None of them are novel in AI litigation, but the combination is aggressive.

• ●Unconstitutional vagueness. xAI argues that core terms in the law — including what counts as a "consequential decision," an "algorithmic discrimination" risk, and a "substantial factor" — are not defined precisely enough to give developers fair notice of what is prohibited, and invite arbitrary enforcement by the Attorney General.
• ●First Amendment (compelled speech). xAI's central argument is that requiring developers and deployers to avoid "algorithmic discrimination" compels model providers to alter their outputs to reflect the state's preferred viewpoints on contested subjects — particularly around race. The carve-outs in the law that explicitly permit outputs that "increase diversity" or correct historical bias are cited as evidence of viewpoint discrimination by the state.
• ●Dormant Commerce Clause — extraterritorial reach. The complaint argues SB 24-205 regulates conduct occurring outside Colorado because AI systems deployed nationally cannot be meaningfully partitioned by state. Complying for Colorado users in practice means complying for everyone.
• ●Dormant Commerce Clause — undue burden. A separate Commerce Clause count argues that even if Colorado has a legitimate interest, the burden on interstate commerce — a 50-state patchwork of AI-specific rules — clearly outweighs the state benefits.

Why the Timing Matters

The case is a near-perfect test vehicle for the legal arguments the December 2025 federal executive order teed up. Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence," specifically named SB 24-205 as an example of a state AI law that, in the administration's view, "may even force AI models to produce false results to avoid a 'differential treatment or impact' on protected groups." That executive order also created the DOJ AI Litigation Task Force, which, as of April 2026, has not yet filed a case of its own against any state AI law.

xAI's private suit runs the same playbook the Task Force was built to run — preemption, dormant Commerce Clause, and First Amendment challenges — but from the private side and without having to wait for DOJ to act. If xAI succeeds in getting a preliminary injunction, it would create a de facto pause on SB 24-205 while the case proceeds. If it loses the preliminary motion, the substantive obligations remain fully enforceable on June 30.

xAI has been through a version of this before. In March 2026, a federal court denied xAI's request for a preliminary injunction against California's training-data transparency law in a separate case. That loss informs the Colorado playbook: xAI is emphasizing compelled-speech and viewpoint-discrimination theories in the Colorado complaint, which tend to trigger higher First Amendment scrutiny than disclosure-focused claims.

Colorado's Other Problem: The ADMT Replacement Framework

Three weeks before xAI filed suit, Governor Polis's AI Policy Workgroup unanimously proposed a new legal framework that would replace the Colorado AI Act rather than amend it. The working group — a mix of industry, consumer, healthcare, and education stakeholders convened by the governor in October 2025 — agreed on the proposal on March 17, 2026. Polis immediately endorsed it.

The draft bill is titled "Concerning the Use of Automated Decision Making Technology in Consequential Decisions." If enacted, it would take effect January 1, 2027, giving covered developers and deployers the rest of 2026 to modify existing compliance programs.

What the ADMT Framework Actually Changes

The proposed framework is not a cosmetic revision. It rewrites the law's footing in several concrete ways.

• ●From "high-risk AI systems" to "Covered ADMT." The new law would regulate automated decision-making technology that processes personal information to generate outputs used in consequential decisions. The scope is closer to California's CCPA ADMT rules than to the original EU-style Colorado AI Act.
• ●Higher trigger threshold. The current CAIA kicks in when an AI system is a "substantial factor" in a consequential decision. The proposal raises the bar to "materially influence," which requires the output to be a non-de-minimis factor that actually affects the outcome.
• ●Several obligations removed. The proposal drops the requirement to report known or reasonably foreseeable algorithmic discrimination risks to the Attorney General, eliminates the mandatory impact-assessment requirement, and removes the mandate to implement a formal risk-management policy (such as NIST AI RMF or ISO 42001).
• ●Consumer protections retained. Coloradans would still get an up-front notice when ADMT is used in life-altering decisions, an explanation for adverse decisions, the ability to correct inaccurate personal information, and the right to request human review.
• ●Liability allocated by relative fault. Developer and deployer liability is split based on who caused the harm. Indemnification clauses that shift a party's own liability to the other are explicitly void as against public policy.
• ●No private right of action, 90-day cure period. Enforcement stays exclusive to the Attorney General. Violators get 90 days to cure before civil penalties attach, unless the violation was knowing or repeated.

Why Polis Supports Replacing His Own Law

Polis signed SB 24-205 reluctantly in May 2024, publicly asking the legislature to "reexamine" the bill before it took effect. In his signing statement he criticized the law for regulating outcomes rather than intent, and raised concerns about Colorado "going it alone" on AI regulation while other states stayed on the sidelines.

The 2025 session produced several failed attempts to amend the law. After a special session pushed the effective date from February 1, 2026 to June 30, 2026, Polis convened the working group in October 2025. The ADMT Framework is the product of that group.

There is also a federal funding angle. The December 2025 executive order on national AI policy conditions access to roughly $21 billion in undeployed Broadband Equity, Access and Deployment (BEAD) funds on the repeal of state AI laws the administration considers "onerous." A replacement law aligned closer to federal preferences is one way to reduce the risk of BEAD-eligibility disputes, regardless of how the Commerce Department's list of "burdensome" state laws eventually lands.

The Two Scenarios That Matter

Between now and the end of Colorado's legislative session on May 13, 2026, two paths are actively in motion, and the outcome for compliance teams depends on which one plays out first.

Scenario 1: The ADMT Framework is enacted. If the legislature picks up the workgroup proposal and passes it before session ends, SB 24-205 is effectively superseded. The June 30 effective date becomes a paper deadline rather than a substantive one, because the repeal-and-replace bill would take over with a January 1, 2027 effective date. Compliance programs then pivot from "high-risk AI system" framing to "Covered ADMT" framing. Impact assessments, AG risk reporting, and NIST-aligned risk-management programs become optional rather than required.

Scenario 2: SB 24-205 takes effect on June 30 as written. If the legislature does not act, or only tinkers at the edges, the current law becomes enforceable on June 30 with its existing obligations intact. In that case xAI's lawsuit becomes the main near-term wildcard. A preliminary injunction would freeze enforcement statewide; a denial leaves the law in force and xAI facing the full statute.

A third possibility is that both happen — the legislature passes the ADMT Framework, and xAI's suit is mooted before it produces a substantive ruling because SB 24-205 has been repealed.

What Deployers Should Be Doing Right Now

None of this changes the practical posture for most organizations preparing for Colorado compliance, and following the headlines is not a substitute for doing the work.

• ●Do not stop your Colorado compliance program. Until the legislature passes the ADMT Framework *or* a federal court enjoins SB 24-205, the current law is on track for a June 30 effective date. A company that deprioritizes Colorado prep based on either the lawsuit or the workgroup proposal is betting on an outcome that has not happened yet.
• ●Keep the work portable. Most of the CAIA compliance artifacts — AI inventory, impact assessments, consumer notices, human review procedures, incident response — also satisfy the ADMT Framework's requirements or come close. Work done for SB 24-205 is not wasted if the ADMT bill passes.
• ●Track the "materially influence" shift. The single biggest scope change between the two laws is the trigger for coverage. Under SB 24-205, almost any AI used in a consequential decision is in scope. Under the ADMT Framework, only systems that "materially influence" the outcome are covered. Once the legislature acts, expect scope reviews to trim a meaningful fraction of previously in-scope systems.
• ●Watch the preliminary injunction motion. If xAI files a preliminary injunction motion (it has not yet, as of filing day), the briefing schedule and hearing date will be the next decisive milestones. A ruling one way or the other before June 30 will substantially change the enforcement picture in the short run.
• ●Do not over-read the First Amendment argument. Even if xAI wins the compelled-speech claim, the most likely remedy is a narrow injunction limiting how Colorado interprets "algorithmic discrimination" with respect to generative output — not the wholesale collapse of the law. Disclosure, documentation, consumer-notice, and human-review obligations are far less exposed to First Amendment challenge.

The Bigger Pattern

Taken together, the two developments tell you more about the state of U.S. AI regulation than either does on its own. The first comprehensive state AI law in the country is under pressure on three fronts simultaneously: the governor who signed it is actively working to replace it, a private company is challenging it in federal court on constitutional grounds, and a federal executive order has publicly named it as a target for preemption litigation.

This does not mean state AI laws are going away. California's TFAIA and SB 942, Texas's TRAIGA, New York's RAISE Act, NYC Local Law 144, and Illinois AIVIA remain in force on their existing schedules. But it does mean that Colorado, which was the first-mover on comprehensive AI regulation, may not be the model the other states ultimately follow. The ADMT Framework, if it passes, will likely become the template for a second wave of state laws that trade comprehensive coverage for narrower, more defensible rules focused on automated decision-making in specific consequential contexts.

The next six weeks — through the close of Colorado's legislative session on May 13 and the original SB 24-205 effective date on June 30 — will determine which version of Colorado's AI law compliance teams are actually working against.

Related Regulations

Sources & References