Colorado's AI Act Nears Its First Real Test

On June 30, Colorado switches on the first broad state law in the country aimed at how AI gets used to make decisions about people. The Colorado AI Act covers what it calls high-risk systems, meaning any AI that plays a substantial role in a consequential decision: a job offer, a mortgage, a place at a school, an insurance quote, access to healthcare or a government service. Developers who build those systems and the companies that deploy them both fall under the rules, as a recent governance review lays out.

The core obligation is to guard against algorithmic discrimination. Deployers have to run documented risk assessments and impact evaluations, keep an eye on systems once they are live, tell consumers when AI factored into a decision that affected them, and stand up a risk-management program rather than treating fairness as an afterthought. The law is built on the idea that a model quietly screening resumes or loan applications can reproduce old patterns of bias at scale, and that someone should have to check.

A deadline that already slipped once

This date was not the original one. Enforcement was meant to begin in February, then got pushed to June 30 after industry argued it needed more time and clearer guidance. That reprieve is now spent. Legal trackers note that the extra months bought companies room to prepare, not an excuse to ignore the law, and the compliance work that looked optional in winter is due now.

Colorado is not alone, which is part of the problem for anyone operating across state lines. Texas brought its own AI law into force at the start of the year, banning systems built for behavioural manipulation or unlawful discrimination. Illinois now requires employers to tell candidates when AI analyses their video interviews. California has moved on labelling AI-generated content. Each state draws its lines a little differently, and a company selling the same tool nationwide has to satisfy all of them at once.

Washington wants the wheel

Hanging over the state-by-state patchwork is a federal effort to flatten it. An executive order signed late last year set up a litigation task force to challenge state AI laws the administration considers obstructive, and told the Commerce Department to weigh which rules conflict with federal policy, with funding used as leverage. It echoes the early-access regime in the more recent June frontier-model order, and runs in the opposite direction from Europe, which has been busy easing its own AI Act.

For now the order changes the politics, not the law. No state statute has been suspended, and a federal executive order cannot by itself overturn one. Until Congress acts or a court rules, Colorado's requirements take effect on schedule. The thing to watch after June 30 is enforcement: whether the state attorney general treats the first months as a grace period for good-faith effort, or starts testing the law on real cases.