
Colorado Scrapped Its AI Act Before It Started. Here Is What Replaces It
Colorado was the first state with a broad AI anti-discrimination law. It never took effect. Two new laws start January 1, 2027, one on automated decisions and one on chatbots. Here is what they require, based on the state's own summaries.
In 2024, Colorado became the first state to pass a broad law aimed at discrimination by high-risk artificial intelligence systems. It was supposed to start in February 2026, then June 2026. It never started at all. In May 2026, the legislature repealed it and passed a narrower replacement, and Governor Jared Polis signed a second law aimed at AI chatbots. Both take effect on January 1, 2027.
If you live or do business in Colorado, this is the AI law you will actually deal with. Here is what the state itself says it does.
From SB 24-205 to SB 26-189
The Colorado Attorney General's office summarizes it plainly: SB 24-205 "created consumer protections from algorithmic discrimination in consequential decisions made by high-risk artificial intelligence systems." SB 26-189, signed in May 2026, "repeals and reenacts those provisions with new requirements regarding the use of automated decision-making technology in consequential decisions."
The General Assembly's bill page lists the signing date as May 14, 2026, after a final 34 to 1 Senate vote on May 12 to accept the House's amendments.
The new law swaps the old term "high-risk AI system" for automated decision-making technology, or ADMT. It covers ADMT used to "materially influence a consequential decision." Consequential decisions are the high-stakes ones in a person's life; legal summaries list areas such as employment, housing, lending, insurance and health care.
| Question | Original SB 24-205 | Replacement SB 26-189 |
|---|---|---|
| Status | Repealed; never took effect | Law; effective Jan. 1, 2027 |
| Core term | High-risk AI system | Automated decision-making technology |
| Developers must | Provide documentation and risk information | Provide technical documentation; notify deployers of material updates |
| Consumers get | Notices and appeal rights (as enacted) | Right to request and correct inaccurate personal data used by ADMT |
| Records | Impact assessments | Retain compliance records at least 3 years |
What Developers and Deployers Must Do
According to the General Assembly's summary, starting January 1, 2027, a developer of covered ADMT must give deployers "technical documentation describing the covered ADMT's intended uses, categories of training data, known limitations, and instructions for appropriate use and human review." Developers must also tell deployers about material updates. Both developers and deployers must keep records showing compliance for at least three years.
The Attorney General adds the consumer piece: the law "gives consumers the right to request and correct inaccurate personal data used by ADMT."
Law firm Krieg DeVault's analysis describes the new law as one that "narrows applicability, eliminates several of the most burdensome obligations" of the original, after "widespread industry concern" about its scope. That is a fair summary of the politics: the replacement is lighter on companies than the 2024 version.
Field note: The original law also became the subject of a federal lawsuit. Policy trackers report that enforcement was stayed in April 2026 in a case brought by xAI. We could not access the court docket directly, so we are flagging it as reported, not confirmed. It does not change the January 1 start of the new law.
The Chatbot Safety Act Rides Alongside
The second law, House Bill 26-1263, is aimed at conversational AI. The Attorney General's office says it requires chatbot operators to estimate users' ages, "disclose that users are interacting with AI and not humans," protect teen users from sexually explicit content and "simulated emotional dependence," and provide privacy and account tools for minors. Operators must also create suicide and self-harm response protocols, and chatbot outputs cannot be presented as equivalent to licensed professional services.
Operators will also file an annual report with the Attorney General. The office says it was signed July 1, 2026, and it also takes effect January 1, 2027.
The Messy Middle: Rules Are Still Being Written
The ADMT law requires the Attorney General's office to adopt rules before January 1, 2027. The office filed proposed rules covering both laws with the Secretary of State on August 11, 2026, released an interim draft on October 6, and held a formal public hearing on October 26. Written comments closed that night. Final rules are the piece to watch over the next six weeks, because they will define terms that the statute leaves open.
Decision Matrix for Colorado Businesses
- If you buy hiring, tenant-screening or lending software: ask the vendor now for the technical documentation SB 26-189 requires, and who reviews decisions.
- If you run a website chatbot: make sure it says it is AI, does not claim to be a licensed professional, and has a path to a human.
- If you build these tools: start the three-year record trail on January 1, not when someone asks for it.
- If you are a consumer: you will have a right to correct inaccurate data used by covered systems. Use it.
The Inflection Point
Colorado's first attempt showed how hard it is to regulate AI in the abstract. Its second attempt is narrower and more practical: tell people when a machine is involved, document what it is for, keep records, and let people fix bad data. Defenders of consumers will say it is too weak; industry will say it is still too much. What matters for regular people is simpler. Starting January 1, 2027, when software helps decide whether you get a job, an apartment or a loan in Colorado, the law expects someone to be able to explain it.
Sources
We checked every fact, figure and quote in this story against the sources below. Quotes are reproduced as published. Images are AI-generated illustrations, not photos of the events described.


