Summary
The Senate Business, Labor and Technology Committee voted to send Senate Bill 4 (the Colorado AI Sunshine Act) to appropriations after supporters cited risks to health, housing and employment and opponents warned of heavy costs and technical infeasibility. The committee adopted four sponsor amendments during the hearing.
The Senate Business, Labor and Technology Committee on Saturday advanced Senate Bill 4, known as the Colorado AI Sunshine Act, after several hours of testimony and multiple sponsor amendments that narrowed parts of the proposal and clarified certain exemptions.
Senator Rodriguez, the bill sponsor, told the committee the measure focuses on transparency and on addressing harms caused when “a computer generated decision and the AI black box is making a decision on health care, education, housing, finance, and essential government services.” He said the bill as revised is “basically a disclosure policy for consumer disclosures for life-changing decisions” and that it is not intended to regulate how AI is built.
The bill would require notice and post-decision information when automated decision systems contribute to consequential decisions — for example in employment, housing, lending, public benefits and healthcare — and would give affected individuals a process to request corrections to the data used in those decisions. Sponsor amendments adopted in committee narrowed the definition of algorithmic decision systems as applied to consequential decisions (L003), added an open-source exclusion (L001), clarified banking and lending disclosure timing and fraud exclusions (L002), and replaced one public-sector disclosure requirement with a route for individuals to request records under the Colorado Open Records Act (L005).
Why it matters: Testimony from consumer groups, civil-rights advocates and health-policy organizations said the bill addresses real risks they say are already occurring. Charles Bridal, director of income and housing policy at the Colorado Center on Law and Policy, cited Medicaid and SNAP problems in other states and a state audit finding that "90% of Medicaid letters" from Colorado’s automated benefits system contained at least one error. Amelia Schachter, a retired oncology nurse testifying for AARP Colorado, said older Coloradans deserve guardrails when algorithmic systems are used in health care.
Supporters argued the bill is a limited, practical set of transparency requirements that will help detect discrimination and give people meaningful ways to challenge adverse, automated decisions. "This bill gives Coloradans notice when automated systems... are used and gives them the right to correct and accurate data that could make the difference between staying enrolled in Medicaid or going without care," Bridal told the committee.
Opponents and groups representing developers, small startups, schools and local governments said the bill, even in its amended form, still sweeps too broadly and imposes technical and legal burdens that will be difficult for small organizations to meet. TechNet and the Colorado Technology Association said the required disclosures — including the proposal for listing the personal characteristics that most influenced a decision — are technically infeasible in many machine‑learning systems and could chill innovation. John Nordmark, a Colorado AI developer, warned the requirements could drive startups out of the state.
Education leaders and school officials expressed particular concern about student and classroom uses of generative tools. Jennifer Prusak of Aurora Public Schools and Michelle Bourgeois, a school district chief technology officer, said teachers and students are using AI for individualized learning and projects; they asked for carve-outs or more practical compliance obligations to avoid deterring classroom innovation.
Several state agencies and business-focused offices registered an amend-and-support or amend position, including the governor’s Office of Information Technology and the Office of Economic Development and International Trade, which said they share the bill’s consumer-protection goals but warned that the law’s broad definitions and joint-liability language raise substantial implementation and legal risks for the state and for small deployers.
Committee action: The committee adopted four sponsor amendments (L001, L002, L003 and L005). The L005 amendment replaced a requirement that public-sector deployers provide a list of the personal characteristics that influenced a decision with a requirement to inform individuals they may obtain that information via a Colorado Open Records Act request. L005 passed on a recorded voice vote in committee and other amendments were adopted without objection. The committee then voted to refer the bill, as amended, to the Committee on Appropriations; the vote passed 4–3.
What’s next: With committee passage the bill goes to the Senate Appropriations Committee. The sponsor said negotiations with stakeholders will continue before floor action. Several opponents urged further technical fixes and clearer scoping before the bill is advanced, while supporters said the revisions struck a balance between innovation and necessary consumer protections.
Ending note: Lawmakers, vendors and advocates agreed at the hearing that AI is evolving rapidly; the central dispute at the hearing centered on how to write disclosure and liability rules that catch consequential uses of automated decision systems without imposing unworkable technical or legal obligations on small businesses, public entities and schools.
