Why Colorado replaced its AI discrimination law with a transparency requirement that the feds might challenge anyway

When Colorado replaced its landmark AI law in May 2026, the move looked like a retreat from ambitious lawmaking. The state abandoned a first-of-its-kind framework that required companies to actively prevent algorithmic discrimination. That’s the risk that automated systems produce biased outcomes based on race, gender, age or other protected characteristics when making decisions about people’s jobs, loans or healthcare.

In its place, on May 14, 2026, the legislature passed something far narrower, including a set of transparency requirements telling companies what they must disclose to consumers.

I study how AI and technology are reshaping policymaking and democratic accountability. I also track state AI legislation through the U.S. State AI Policy Tracker at the University of Denver’s Daniels College of Business.

Colorado’s back-and-forth on AI legislation exposes a fundamental shift in how state leaders are thinking about governing AI. It also raises an unresolved constitutional question that could determine whether any state AI law survives federal challenge.

From prevention to disclosure

The original Colorado law, called the Colorado Artificial Intelligence Act, asked companies to prevent harm. The replacement asks companies to inform consumers. That distinction may sound technical, but it reflects a meaningful change in governing philosophy.

The Colorado Artificial Intelligence Act was built on direct obligations. Developers and deployers of high-risk AI systems – such as tools used to screen job applicants, evaluate loan applications or determine insurance eligibility – had to establish risk management programs. The law also required companies to conduct impact assessments and take proactive steps to avoid discriminatory outcomes. In essence, the state was telling companies: Govern yourselves, and prove you’re doing it.

The replacement law, called the Automated Decision-Making Technology Act, takes a different approach.

Companies must notify consumers when automated technology plays a role in a consequential decision. That could include a hiring determination or a loan denial. If that decision goes against a consumer, the company has 30 days to explain how the technology contributed to the outcome.

Consumers can also request corrections to inaccurate personal data and ask for meaningful human review. The duty of care, risk management mandates and impact assessments from the first iteration of the law are gone.

For Coloradans, the practical difference is significant. Under the old law, a company using AI to deny a loan application, for example, would need internal processes designed to catch discriminatory patterns before they reached the consumer.

Under the new law, that company must tell the consumer that AI was involved and explain its role after the fact. However, the company has no obligation to audit the system that produced the decision. The burden shifts from the institution deploying the technology to the individual affected by it.

Equal protection out, free speech in

A federal lawsuit filed in April 2026 brought the timeline for this legislative overhaul forward quickly. Elon Musk’s company xAI sued to block enforcement of the Colorado Artificial Intelligence Act, the original law.

Days later, the DOJ intervened in support of xAI. It was the first time the federal government had moved to invalidate a state AI law.

In its legal filing, the DOJ made two constitutional arguments against the original Colorado law. The new law neutralized the first argument but may have sharpened the second.

The first was an equal protection argument. The DOJ argued that the law’s antidiscrimination framework effectively forced developers to make race- and sex-conscious decisions about how their models behave. For example, companies would have been required to test whether their AI systems produced different outcomes for different demographic groups and to adjust them if they did.

That kind of government-imposed classification is one courts subject to heightened scrutiny, meaning the government bears a heavy burden to justify it.

The new law eliminated this argument entirely. The algorithmic discrimination provisions are gone, and there is nothing left in the statute that requires developers to monitor or adjust their outputs for discriminatory patterns.

The second argument made by the DOJ was a First Amendment argument, specifically what is called compelled speech. The idea is straightforward: The First Amendment does not just protect a person’s right to speak, it also protects the right not to speak. When a government requires a private company to deliver specific messages to consumers, that mandate has to meet a constitutional standard. Courts ask whether the government has a strong enough reason to force someone to say something they otherwise would not.

xAI characterized the Colorado law’s disclosure and reporting requirements as exactly this kind of forced communication.

In my opinion, this argument was strengthened, not weakened, by the rewrite. The new Automated Decision-Making Technology Act kept little besides notice and disclosure duties, so the new law is, at its core, the government telling companies what they have to say to consumers.

In narrowing the statute this way, the state legislature may have made the compelled speech argument easier to isolate and litigate, even as it made the equal protection argument disappear.

The executive order behind the lawsuit

The DOJ’s decision to intervene in a state-level lawsuit did not come out of nowhere. It followed directly from Executive Order 14365, which President Donald Trump signed in December 2025.

The order declared that a “minimally burdensome national policy framework” should govern AI, and it directed the DOJ to challenge state laws that conflict with that vision. The order also instructed the Commerce Department to publish an evaluation of existing state AI laws, identifying those it deems “onerous.” It included a specific mandate to flag laws compelling disclosures that could violate the First Amendment.

That evaluation was due by March 2026, but it has not yet appeared. If that list materializes – and if Colorado’s new law lands on it – it will signal that the administration sees this state not as a single problem to resolve but as a test case for a broader federal campaign against state AI regulation.

Dozens of states have introduced AI-related legislation in 2026 that touches on the same areas Colorado tried to regulate, including employment, lending, housing and healthcare.

What Coloradans should expect

For now, not much changes in Colorado. A federal judge stayed enforcement of the original law in April 2026, and that stay applies to the replacement law as well. This means none of its provisions can be enforced yet.

The stay will remain in place until at least 14 days after the court rules on xAI’s request for a preliminary injunction, which is a court order that would block the law from taking effect while the case is decided. That request will not be filed until 28 days after the state completes its rulemaking process. So the timeline depends on the attorney general, the courts and the DOJ’s willingness to press a challenge against a law that already conceded substantial ground.

Read more of our stories about Colorado.