Colorado Just Gutted Its Own AI Law Before It Could Take Effect

By Hector Herrera | June 6, 2026 | Government · Policy

Colorado's Governor Jared Polis signed SB 189 in May 2026, fundamentally rewriting the state's AI Act and pushing its enforcement date from June 30, 2026 to January 1, 2027. More significantly, the revision strips out the law's core regulatory mechanisms — the provisions that made it the most ambitious state AI legislation in the country — and replaces them with softer disclosure requirements that critics say lack teeth.

The original Colorado AI Act, passed in 2024, had required developers and deployers of "high-risk AI systems" to conduct algorithmic impact assessments, implement bias-testing protocols, and provide consumers with the right to appeal AI-driven decisions affecting employment, housing, education, and healthcare. SB 189 eliminates the duty-of-care framework and mandatory impact assessments entirely, replacing them with narrower disclosure requirements around automated decision-making that do not include enforcement mechanisms comparable to what was removed.

A Significant Retreat

To understand how significant this is, you need to understand what the original Act had promised.

Colorado's 2024 law was modeled partly on the EU AI Act and partly on Colorado's own Consumer Protection Act. It treated high-risk AI systems — those making or substantially influencing decisions about individuals — like products with inherent liability exposure. Developers had an obligation to design systems that didn't produce discriminatory outcomes. Deployers had an obligation to audit and document. Consumers had a right to human review when AI decisions went against them.

That framework is now gone. What remains is essentially a disclosure-only regime: companies must tell consumers when automated decision-making is involved and provide a general contact point for questions. There are no mandatory audits, no algorithmic impact assessment requirements, and no codified standard of care for high-risk AI deployments.

Who Won and Who Lost

Tech industry groups won. The Colorado Technology Association, the Chamber of Progress, and national trade groups representing AI developers lobbied aggressively against the original Act's impact assessment and bias-testing requirements, arguing they would impose compliance costs that made Colorado hostile to AI development and that they imposed requirements without clear technical definitions of compliance.

Consumer and civil rights organizations lost. Groups including the Colorado Center on Law and Policy and the National Fair Housing Alliance had supported the original framework as a meaningful check on automated decisions in housing and employment — two of the areas where algorithmic bias is best documented. The new disclosure requirements do not give consumers meaningful recourse when AI decisions produce discriminatory outcomes.

Governor Polis, a Democrat and self-described tech optimist, has consistently argued that Colorado should lead on responsible AI development rather than restrict it. His decision to sign the rollback over objections from civil rights advocates reflects a calculation that the original law's requirements were so broad that they risked being unworkable and that the compliance ecosystem to support them doesn't yet exist.

What the New Law Actually Requires

The amended Act's disclosure requirements center on two obligations:

1. Notification: When a covered entity uses automated decision-making to make a consequential decision about a consumer (employment, housing, credit, insurance, education), the consumer must be notified that automated systems were involved.

2. Contact information: The consumer must be provided with a way to reach a human representative to ask questions about the decision.

There is no requirement that the human representative have authority to override the AI decision, and no mandatory process for the company to document how the automated system works or how it was validated.

Why This Sets a Cautionary Precedent

Colorado was supposed to be the template. After the EU AI Act's 2024 passage, American advocates pointed to Colorado's law as evidence that meaningful AI governance was achievable at the state level without waiting for federal action. A dozen states were watching Colorado closely before drafting their own proposals.

SB 189 sends a different message: that even states with the political will to pass comprehensive AI legislation can be walked back by industry pressure before enforcement ever begins. It demonstrates that the disclosure-only approach — which the tech industry consistently argues is sufficient — can be imposed on more ambitious frameworks by sustaining enough lobbying pressure in the implementation phase.

The Colorado AI Act was amended the week before its original enforcement date. That is not a normal policy revision timeline. It reflects how much was at stake for industry groups who had seen the June 30 deadline approaching without a viable compliance pathway.

What to Watch

Other state legislatures are now recalibrating their approaches. States that were drafting comprehensive AI bills with impact assessment requirements have seen Colorado's rollback and are assessing whether those provisions are politically sustainable. Expect some states to scale back proposals before introduction.

Federal preemption is the backdrop. The White House's push for a federal AI governance framework that would preempt state laws is gathering momentum — and is explicitly designed to set a national floor that prevents states from imposing requirements as stringent as Colorado's original Act. If federal legislation passes in 2026, state retreats like Colorado's become legally locked in.

January 2027 is the new enforcement date for even the weakened disclosure requirements. Whether Colorado's attorney general actually enforces those requirements will be the first real test of whether the revised Act has any practical effect.

Colorado's experience is a case study in how AI policy can be passed and then effectively neutralized — not by repeal, but by amendment. The lesson for advocates drafting the next round of state AI legislation is to build in protections against exactly this kind of late-stage rollback.