First US state law against algorithmic discrimination by AI gutted before taking effect - contrast to EU AI Act
What it really says
On June 30, 2026 - tomorrow - the Colorado AI Act was supposed to take effect as the first comprehensive AI law of any US state. The law would have required developers and deployers of high-risk AI systems to protect consumers from algorithmic discrimination in areas such as employment, education, financial services, healthcare, insurance, and housing. But Governor Jared Polis signed the replacement law SB 189 on May 14, 2026, repealing the original Colorado AI Act just 47 days before its enforcement date and replacing it with a significantly weaker framework. What was removed: the duty of care to prevent algorithmic discrimination, mandatory impact assessments, risk management programs, and certain reporting obligations to the attorney general. What remains: consumers have a right to disclosure when automated systems are used in consequential decisions, as well as a right to meaningful human review of adverse outcomes. The new law SB 189 will not take effect until January 1, 2027. Consumer advocacy groups have criticized the change as gutting the law. The technology industry had argued the original law was too vaguely worded and would impose disproportionate documentation costs on small and medium-sized businesses.
Our assessment
This development merits a yellow rating because it shows how difficult it is for the US to implement binding AI regulation, which has both reassuring and concerning aspects. The legitimate concern: when even the most ambitious AI law of any US state is so heavily watered down before taking effect, it raises the question of whether democratic processes can keep pace with the speed of AI development. Removing the duty of care against algorithmic discrimination means US consumers continue to have little legal protection when AI systems make discriminatory decisions in lending, hiring, or insurance. The other side: not every regulation is automatically good regulation. The original law did have definitional problems, and overly strict requirements can hamper innovation without effectively protecting consumers. The remaining transparency right, that consumers learn when AI is used in important decisions, is a reasonable first step. The right to human review of adverse outcomes also remains. The question is whether transparency alone is sufficient when the actual duty to prevent discrimination is missing.
Relevance for Germany
This development is particularly instructive for Germany and Europe because it highlights the fundamental difference between the US and European approaches to AI regulation. While Colorado weakened its law before it could take effect, the EU AI Act is approaching full applicability in August 2026 with significantly stricter requirements. The EU pursues a risk-based approach with binding duties of care, conformity assessments, and fines of up to two percent of global annual turnover. The failed Colorado law was supposed to be more similar to the EU approach. Its failure could strengthen the position of those in the EU defending the stricter European path: if voluntary or market-driven regulation does not work, binding rules are needed. At the same time, critics in Europe also warn that overly strict regulation could disadvantage European companies against less regulated competitors from the US and China. Germany's AI Market Surveillance Act, passed by the Bundestag on June 11, designates the Federal Network Agency as the central supervisory authority for the EU AI Act in Germany, a concrete step toward implementing the European approach.
Fact check
Information about the Colorado AI Act and its replacement by SB 189 comes from the official Colorado General Assembly website, where the legislative text and voting records are publicly accessible. Governor Polis's signing on May 14, 2026, is documented. The analysis of the legislative changes is consistently confirmed by multiple independent legal analyses, including Holland & Knight, Davis Wright Tremaine, Wilson Sonsini, Seyfarth Shaw, and Norton Rose Fulbright. The IAPP confirms the shift from a risk-based to a transparency-based approach. Criticism from consumer advocacy groups is documented in TechTimes reporting. Germany's AI Market Surveillance Act passage on June 11, 2026, is reported by Heise Online.
Source
- • https://leg.colorado.gov/bills/sb26-189
- • https://iapp.org/news/a/amendments-move-colorado-ai-act-s-focus-from-risk-to-transparency
- • https://www.techtimes.com/articles/318956/20260623/colorado-ai-law-hits-june-30-deadline-without-bias-audits-here-what-consumers-keep.htm
- • https://www.hklaw.com/en/insights/publications/2026/05/colorado-governor-signs-sb-189