Colorado Governor Jared Polis expressed "reservations" when signing the state’s artificial intelligence law (SB 24-205), criticizing its complexity and potential negative impact on technological advancements. He advocated for amendments to align the Act with evidence-based recommendations, prompting his AI Policy Working Group to release a proposed bill. This draft legislation aims to significantly ease compliance burdens for employers by eliminating many onerous obligations, although it introduces new requirements and broadens the scope of regulated technology, indicating a dynamic legislative process ahead.
Proposed Eliminations of Employer Obligations
The draft bill aims to significantly reduce the compliance burden for employers by removing several key mandates from the existing Act. These include requirements for reporting discriminatory outcomes to the Colorado attorney general, conducting comprehensive impact assessments of their AI systems, and implementing a comprehensive risk management policy and program. The mandates for annual reviews of AI tools and the need to post or update privacy policies specifically describing AI tool usage would also be eliminated. Furthermore, requirements to provide notice when interacting with an AI system and to comply with specific correction requests would be removed. Importantly, the bill would move away from affirmatively avoiding algorithmic discrimination, instead relying on compliance with existing state and federal anti-discrimination laws, streamlining the regulatory framework considerably for employers utilizing artificial intelligence in employment decisions.
Retained and New Obligations for Employers
Despite significant removals, the proposed bill introduces a refined set of obligations and broadens certain aspects of the regulation. It preserves the critical transparency requirement for employers to provide notice either before or concurrently with the use of a covered AI tool, ensuring individuals are aware of AI involvement. Post-adverse action notices would also remain mandatory. Deployers of AI tools would still need to provide detailed information about the systems they develop. The bill introduces new record-keeping duties, requiring employers to retain records of their covered AI tool usage for a period of three years. A significant change is the substantial expansion of the definition of "covered technology." Under the new proposal, any computational process that uses personal information and materially influences employment or other consequential decisions would fall within the law’s purview, extending beyond the previous, narrower definition of "artificial intelligence systems." Although some clerical uses are carved out, this expansion could encompass a broader range of automated processes not previously considered AI. Lastly, the proposal raises potential complexities by suggesting an extension of Colorado Privacy Act obligations, currently limited to consumer data, to include applicant and employee data, which could introduce new compliance challenges.
Future Legislative Process and Effective Date
The proposed bill, still a draft from the Governor’s AI Policy Working Group, is slated for significant evolution throughout the upcoming legislative process. It is highly anticipated that this draft will undergo substantial revisions, and there remains a possibility that state lawmakers may introduce alternative or competing bills, mirroring legislative dynamics observed during the August 2025 special session. Should a revised version of the Act ultimately be enacted based on this proposal, it would provide a much-needed reprieve for employers to adapt to the new regulatory landscape. The effective date for the new law would be deferred until January 1, 2027, offering a more extended period for compliance preparation compared to the original February 1, 2026, and subsequently extended June 30, 2026, deadlines. This timeline adjustment underscores the state's recognition of the complexities involved in AI regulation and the need for businesses to adequately prepare.
