Colorado Delayed Its AI Hiring Law Twice — Don’t Read That as “We Can Wait”

Colorado’s AI Act was supposed to take effect February 1, 2026. It didn’t. Governor Polis signed a bill pushing that date to June 30, 2026. Then, on May 14, 2026, he signed SB189, delaying it again to January 1, 2027, and stripping out a chunk of the original law’s teeth along the way. Two delays and a rewrite in under a year, for a law that specifically targets AI use in hiring decisions.

If you run TA or people-ops for an Indian company that screens candidates for US-facing roles, or your GCC parent has US hiring exposure, the natural reading of “delayed twice, scope narrowed” is “we have time.” That reading is wrong, and it’s wrong for a reason worth naming plainly.

What actually got cut, and what didn’t

The original Colorado AI Act (SB 24-205) imposed a duty of care on employers using AI in hiring to prevent algorithmic discrimination, required deployer risk-management programs and impact assessments, and created reporting obligations to the state Attorney General. SB189 removed most of that. What’s left, per the Colorado Sun’s coverage of the rewrite, is a narrower law built around disclosure and transparency: telling candidates AI is involved in a decision about them, and giving them a way to understand or contest it.

That’s a real scale-back. It’s also not nothing. Disclosure-and-transparency requirements are the same core shape as Illinois’s HB3773, which we’ve covered before, and they’re the same core shape as what the EU AI Act asks of high-risk hiring systems. Colorado’s legislature couldn’t agree on the harder half (impact assessments, discrimination liability), but it kept the easy-to-explain half: candidates get told when AI is in the loop, and a human stays reachable if they push back.

Why the timeline isn’t the thing to track

Two delays in one year should tell you something about how unstable state-by-state AI hiring law is right now, not that the underlying pressure is going away. Colorado’s law is one line in a longer pattern: Illinois’s disclosure requirement is live, California has finalized new AI employment-discrimination regulations in the same window Colorado was rewriting its law, and NYC’s Local Law 144 has required bias audits for automated hiring tools since enforcement began in July 2023. Any one of these can slip its effective date. The direction hasn’t slipped once.

For an Indian TA team, the practical exposure isn’t “does Colorado’s law apply to us specifically.” It’s that global clients running vendor and compliance due diligence increasingly ask a version of the same question regardless of which state’s law is currently in force: what does your AI screening actually do, can a candidate find out, and is there a human in the loop before a rejection is final. A GCC or BPO answering “we’re waiting to see which US law lands first” to that question is answering a compliance officer’s question with a legislative-tracking answer. Those aren’t the same thing.

The contrarian take: treating AI-hiring compliance as a state-law-tracking exercise guarantees you’re always reacting to the slowest-moving target instead of the fastest one. The laws that keep getting delayed and rewritten are not where the real requirement is settling. The requirement that’s held steady across every version, Colorado’s original draft, Colorado’s rewrite, Illinois, and the EU AI Act, is disclosure plus a human override. That part isn’t going to get repealed in the next rewrite.

What this looks like in practice, not in a legal brief

You don’t need outside counsel to start on the part that’s stable. A candidate should be able to find out, in plain language, that an AI conversation was part of their screening. A rejection produced by that screening should route to a status a human reviewed, not an automatic close. We built HireQwik’s decision output as a four-way call (Strong Go, Go, On Hold, No Go) with a human step in the loop for exactly this reason, not because Colorado required it on any of its three effective dates, but because every version of every AI-hiring law we’ve tracked has converged on wanting that same thing from a different angle.

The honest limit

This isn’t a call to over-build for a regulatory environment that keeps rewriting itself; Colorado’s own legislature spent a year failing to agree on its own law, twice. The lesson isn’t “assume the strictest possible version.” It’s narrower than that: disclosure and human override are the two requirements that have survived every rewrite so far, in every jurisdiction we’ve tracked. Build for those two, and the next delayed, narrowed, or rewritten state law is a formality, not a scramble.

If you want to see what a disclosure-ready, human-in-the-loop screening flow actually looks like before you need it for a client audit, talk to us.