Today, June 30, 2026, was supposed to be the day American employers finally had to answer for the algorithms screening their job applicants. Colorado’s Artificial Intelligence Act — the first comprehensive state law in the country to make companies actively check their AI hiring tools for discrimination — was scheduled to take effect this morning. Instead, it is already gone. Six weeks ago, with one day left in the legislative session, lawmakers passed a replacement, and the governor signed it. The version that survives is a much softer thing, and it does not arrive until January 2027.
What the old law would have demanded
The original Act, SB24-205, treated AI in employment as something to be audited. Any employer using AI as a substantial factor in hiring, pay, or promotion would have had to run impact assessments, stand up a formal risk-management program, review its tools annually for discriminatory outcomes, report any bias it found to the state attorney general, and — the load-bearing clause — take affirmative steps to avoid algorithmic discrimination in the first place. It was demanding, it was unpopular with employers, and it was, for a brief window, real.
That last duty was the one that mattered. “Affirmatively avoid algorithmic discrimination” puts the burden on the company to prove its black box is not quietly filtering out older applicants or sorting by zip code. Take it away and you are left with the right to complain after the fact, which is the right workers already had.
What replaced it
The new law, SB26-189, keeps the vocabulary and removes the teeth. According to an analysis by employment firm Littler, the rewrite deletes the impact assessments, the risk-management program, the annual reviews, the duty to report discriminatory outcomes, and the affirmative obligation to avoid bias. What remains is procedural: tell people you are using an automated tool, give them a notice within 30 days if it contributes to a decision against them, offer a “right to correct” their data and a “meaningful human review” — but only, the statute says, “to the extent commercially reasonable.” Keep the records for three years.
“To the extent commercially reasonable” is the kind of phrase that sounds like a safeguard and functions like an exit. As Littler dryly notes, if building a parallel human-review process costs more than the automation saves, an employer has “a reasonable argument that compliance is not commercially reasonable.” The escape hatch is written into the protection.
Who can do anything about it
Almost no one. Like its predecessor, the law creates no private right of action — a rejected applicant cannot sue. Enforcement belongs solely to the Colorado attorney general, who must first hand violators a 60-day notice and a chance to cure before penalties of up to $20,000 per violation apply. One office, policing every AI hiring decision in the state, after a cure period, with no help from the people actually screened out.
The timing is the part worth sitting with. Colorado was the third US jurisdiction, after California and New York City, to write comprehensive rules for AI in employment. Its law was weakening at the exact moment the labor market gave everyone the clearest reason yet to want one: AI-attributed layoffs cleared 150,000 in the first half of 2026, and the same screening systems now stand between displaced workers and their next job. The one rule that would have forced a look inside the machine got hollowed out in an empty committee room, six weeks before the public would have noticed it was supposed to exist.
For job seekers, nothing visible changes today. The automated resume filter still runs, still rejects, still explains nothing. The difference is that for about six weeks this spring, there was a law on the books that said it had to do better — and now there is a lighter one that says it mostly has to send a memo.
