State efforts to regulate AI decision-making tools are veering away from bias audit mandates toward transparency requirements, pushed along by Colorado’s narrowing of its bellwether algorithmic bias law.
The revised law (SB 189), which Gov. Jared Polis recently signed, will require employers to give workers notice about the artificial intelligence tools they’re using to guide their hiring, promotions, and firing, much like a Connecticut bill (SB 5) awaiting that governor’s signature.
The two states join California and New York City in requiring varying degrees of disclosure to employees and job seekers when bosses use AI for employment decisions, as well as Illinois where regulators just issued proposed rules to flesh out an existing notice-focused law.
Colorado’s move to replace its broadest-in-the-nation algorithmic bias law (SB 205) on May 14, just weeks before its effective date, aligns with a wider state-level pivot away from AI decision-making bills that prescribe specific anti-bias practices like audits. White House efforts to preempt some state AI laws helped spur legislatures to scale back to simpler disclosure mandates.
“It seems that legislatures have coalesced around this idea that the best way to protect people is to let them know what you’re doing” with AI decision tools, said Jennifer G. Betts, an office managing shareholder at Ogletree Deakins whose practice includes advising employers on AI use.
Pressure from the tech industry to avoid stricter AI regulation also helped fuel the shift toward transparency requirements. The Chamber of Progress, whose members include
It’s unclear how effective disclosure-oriented regulatory efforts will be at preventing unseen discrimination facilitated by AI tools based on race, sex, disability, or other protected traits. Colorado lawmakers said their revised law aims to let workers and consumers know the criteria that go into automated decisions and correct inaccurate data, but without blocking innovation in the state.
“This new law strikes an appropriate balance of protecting consumers while not being onerous on developers or the businesses who use AI technology,” Colorado state Senate Majority Leader Robert Rodriguez (D), who sponsored the legislation, said in a statement.
Complying with the revised Colorado measure, which largely takes effect Jan. 1, 2027, will be considerably less onerous for employers than SB 205 would have been, employment and AI-focused attorneys said. But exactly how onerous will depend on the details of regulations to be implemented by the state’s attorney general and how courts interpret the statute.
Colorado’s notice requirements will still be broader than other states’ in several ways. For one, existing measures focus largely on use of AI in employment-related decisions. The Colorado law covers a range of “consequential decisions,” including those affecting people’s access to education, financing, government services, healthcare, housing, and insurance.
The state also will require giving workers and consumers the chance to correct inaccurate data and request meaningful human review “to the extent commercially reasonable” when AI-powered decisions reject them.
Bias Prevention
Colorado’s attorney general has sole enforcement authority over the disclosure requirements, similar to the Connecticut bill. The revamped law provides no private right of action, but workers still have the ability to bring claims under existing anti-discrimination laws.
AI technology is so ubiquitous and built into systems people interact with every day that requiring a notice to tell people they’re interacting with AI is unlikely to provide much consumer protection, said Adam Klein, managing partner at plaintiff-side law firm Outten & Golden LLP.
Even the companies using the technology don’t understand how it operates well enough to know whether it’s discriminating, he said. Nevertheless, gradually unveiling more information about the way AI decision tools work could lead to companies being held accountable when systems promote discrimination or violate workers’ rights.
“There’s probably going to be a watershed moment where enough information will be compiled that there’s going to be a reckoning,” Klein said.
While California’s civil rights and privacy agencies each passed regulations in 2025 focusing on transparency disclosures when businesses use decision-making tools, the civil rights rules include language some lawyers described as a “backdoor mandate” for audits. The rules say employers facing AI-related discrimination claims can use evidence that they audited their tools in defending themselves.
The Connecticut bill also says bias audits can serve as a defense but doesn’t require them.
Open to Interpretation
Parts of the Colorado law could be narrowed or clarified through the AG’s rulemaking or court decisions, said Cody Barela, a technology-focused partner at Armstrong Teasdale LLP in Denver.
One point he said that’s open to interpretation: how to determine whether a tool “materially influences” consequential decisions and thus triggers the law’s notice requirements.
The makers of some hiring and recruiting technologies “might say the intended purpose is just to organize, just to help in certain aspects, but the tool is not intended to have any meaningful decision-making in the outcome,” Barela said. “That might be a way some developers try to get out of some of the obligations.”
The caveat that businesses only have to grant meaningful human review when commercially reasonable also could limit employers’ compliance burdens and workers’ rights.
Particularly for large businesses receiving thousands of applications, providing human review for each rejected candidate might not be feasible, Barela said.
If nothing else, the passage of a revised Colorado law likely helps the state escape the pending legal challenge by Elon Musk’s xAI and the US Justice Department that led to a halt on enforcement of the previous law, said Charlie Bullock, senior research fellow at the Institute for Law & AI.
A central claim in the litigation was that SB 205 infringed on AI developers’ First Amendment rights by requiring them to encode pro-diversity preferences into their technology. The diversity language they contested is absent from the new law.
“There’s really nothing to litigate now,” Bullock said.