Colorado's AI Act: What Must Be in the Consumer Notice

Colorado’s AI Act, SB 24-205, codified at C.R.S. §§ 6-1-1701 to 6-1-1707, imposes three separate disclosure duties on a deployer of a high-risk AI system. They have different triggers, different audiences and different content, and treating them as one privacy-policy paragraph will satisfy none of them.

Three notices, not one

The deployer duties sit in § 6-1-1703. Within it, one notice is owed to a consumer before a high-risk system makes or is a substantial factor in making a consequential decision about them; another is owed after such a decision, but only where the decision is adverse to the consumer; and a third is a standing public statement on the deployer’s website. A fourth duty, in § 6-1-1704, is unrelated to high risk entirely: it requires disclosure whenever a consumer interacts with an AI system at all.

The trigger vocabulary is doing work. A “consequential decision” is one that has a material legal or similarly significant effect on the provision or denial to a consumer of, or the cost or terms of, education enrolment or an education opportunity, employment or an employment opportunity, a financial or lending service, an essential government service, health-care services, housing, insurance, or a legal service. Note that the definition reaches cost or terms, not only provision or denial — so a pricing model in insurance or lending is caught even where nobody is refused. And “substantial factor” means a factor that assists in making the decision and is capable of altering its outcome, which is a much lower bar than the sole-automation tests used in some privacy statutes.

The pre-use notice

Before, or at the time of, deploying a high-risk system to make or be a substantial factor in a consequential decision, the deployer must notify the consumer of that fact, and provide:

• a statement disclosing the purpose of the high-risk AI system and the nature of the consequential decision;
• the deployer’s contact information;
• a plain-language description of the system — a description aimed at the consumer, not the vendor’s datasheet;
• instructions on how to access the deployer’s public statement about the high-risk systems it deploys; and
• information about the consumer’s right to opt out of profiling in furtherance of decisions that produce legal or similarly significant effects, under the Colorado Privacy Act.

That last field is the one people miss, and it is a cross-reference rather than a new right: the Colorado AI Act tells the deployer to point the consumer at the profiling opt-out that already exists in C.R.S. § 6-1-1306(1)(a)(I)(C). The two statutes are therefore coupled, and a deployer’s AI notice has to be accurate about the privacy right as well as about the model. The notice must be delivered in plain language, in all languages in which the deployer ordinarily does business, and in a format accessible to consumers with disabilities.

The adverse-decision notice

Where the high-risk system makes, or is a substantial factor in making, a consequential decision that is adverse to the consumer, a second and much more demanding notice is owed. It must disclose:

• the principal reason or reasons for the decision, including the degree to which, and the manner in which, the high-risk AI system contributed to it. Not that a model was used — how much it mattered and how;
• the type of data processed in making the decision and the source or sources of that data; and
• an opportunity to correct any incorrect personal data the system processed, and an opportunity to appeal.

Read the first bullet as an engineering requirement rather than a drafting one. Producing a principal-reasons statement that describes the degree and manner of the model’s contribution means you need to know, per decision, which features drove the output and what the human decision-maker did with it. Adverse action reasoning is familiar territory in consumer credit — see adverse action notices for AI credit denials — but Colorado extends the same logic to employment, housing, education, insurance, health care, legal services and essential government services, most of which have no equivalent existing practice.

Correction and human appeal

The appeal right is qualified in a way worth quoting precisely: the deployer must provide an opportunity to appeal an adverse consequential decision, which appeal must, if technically feasible, allow for human review. The qualifier is real and it is also a trap. A deployer asserting that human review is not technically feasible is making a claim about its own architecture that it chose, and it will be judged against what similar deployers manage. Building a system in which human review is impossible is not obviously a defence to a requirement that human review be provided where it is possible.

The correction right is narrower than it sounds: it attaches to incorrect personal data the system processed, not to the model’s inference. A consumer can tell you that your record of their income is wrong; the statute does not give them a route to argue that the score was wrong on correct data. The appeal is the route for that.

The public statement and the interaction disclosure

Separately from anything owed to an individual, a deployer must make available on its website a statement summarising the types of high-risk AI systems it currently deploys, how it manages known or reasonably foreseeable risks of algorithmic discrimination arising from them, and the nature, source and extent of the information it collects and uses. The statement must be updated as the deployment changes. It is a standing public artefact, which means it is also the first thing a plaintiff’s lawyer or the Attorney General will read, and the first place an inconsistency with your internal impact assessment will show.

§ 6-1-1704 adds the general one: a developer or deployer of an AI system intended to interact with consumers must ensure the consumer is disclosed that they are interacting with an AI system, unless it would be obvious to a reasonable person. That duty does not depend on high risk, on a consequential decision, or on anything else — it applies to an ordinary support chatbot. It is the Colorado analogue of California’s bot disclosure law and of the EU AI Act’s Article 50 duty, and it is broader than either, because it has no commercial-purpose limitation.

Two operational points close this out. Enforcement is exclusively by the Attorney General — there is no private right of action, and a violation is treated as a deceptive trade practice under the Colorado Consumer Protection Act. And the Act’s application date was moved from 1 February 2026 to 30 June 2026 by legislation in the August 2025 special session, so any notice programme built to the original date has more room than it thinks. The effective-date page tracks that, and the liability shields available to a deployer who does all of this are in the rebuttable-presumption page.