Colorado’s Proposed Rules Require Meaningful Human Review And Reconsideration Of Employment Decisions Materially Influenced By AI

Duane Morris Takeaways:  On August 11, 2026, the Colorado Department of Law released a set of proposed rules intended to govern the implementation of Colorado’s Automated Decision-Making Technology Act (“ADMT Act”) and Conversational Artificial Intelligence Service Operator Requirements (the “Chatbot Safety Act”). Both laws, and the proposed rules (to the extent they become final, following a notice and comment period), take effect January 1, 2027.

The proposed rules under the ADMT Act would add additional layers of regulatory burdens on companies using AI-based tools to make consequential automated decisions affecting a consumer’s access to, eligibility for, or terms of employment, education, housing, lending, financial services, insurance, health care, essential government services, and public benefits. 

Take AI employment tools, for example — if Colorado’s ADMT rules were to become final, they would require companies using AI employment tools to make available and readily accessible to job candidates and employees:

1. the make, model, and version number of any AI tool whose outputs were used or will be used to materially influence a consequential employment decision such as in candidate screening, interviewing, hiring, firing, and other consequential employment decisions;
2. the types and categories of personal data input to the model;
3. the sources of these input data, including the identities of intermediary and original sources; and
4. to the recipients of any adverse outcomes materially influenced by the AI tool:
   1. within 30 days, notifications of the adverse outcome setting forth the decision, purpose of the AI tool, the reasoning and primary factors relied upon by the AI tool to produce the output, and the relative roles of the AI tool and human reviewers in the decision — such as whether the employer followed the human best practices to mitigate the risk of AI hiring tool noncompliance with antidiscrimination statutes that we identified in our blog (here); and
   2. within 45 days after receipt of a request for review, complete a meaningful human review and reconsideration of any adverse outcome, to the extent commercially reasonable.

The proposed rules under the Chatbot Safety Act would add age-assurance, disclosure, privacy, minor-protection, crisis-response, and reporting requirements on companies using AI-based chatbots.

The analysis in our blog post focuses on the “meaningful human review and reconsideration” component of the ADMT Act and its corresponding proposed rules, in the context of adverse employment decisions (number 4(b), above), to assist companies in preparing for compliance by January 1, 2027, and in commenting on the proposed rules before the comment period closes on October 26, 2027, as may be appropriate for their businesses. 

Organizations evaluating and seeking to comply with the proposed rules should continue monitoring the rulemaking process, as key provisions, including those discussed below, may change before the rules become effective.

Background

Colorado’s ADMT Act (Colo. Rev. Stat. § 6-1-1701, et seq.) was enacted May 14, 2026, and will become effective January 1, 2027. It provides, among other things, that when an employee or job applicant experiences an adverse outcome resulting from a consequential decision that was materially influenced by AMDT such as an AI employment tool, then the employee or job applicant may request, and the employer must provide (i) instructions for requesting personal data and correcting factually incorrect or materially inaccurate personal data used in the decision; and (ii) “an opportunity for meaningful review and reconsideration of the consequential decision, to the extent commercially reasonable.” Id. § 6-1-1705. 

“Meaningful Human Review,” as defined under the ADMT Act, requires the employer to designate an individual who has authority to approve, modify, or override a consequential decision to perform the review. Id. § 6-1-1701(15). Further, in conducting the review, the designated individual must consider relevant, available primary evidence, must be trained to conduct the review, must not default to the system output, and must have access to sufficient information to understand (i) the output’s intended use, material limitations, and categories of inputs; and (i) the principal factors used to generate the output, without requiring disclosure of proprietary source code, model weights, or other trade secrets. Id.

The ADMT Act also provides that, on or before January 1, 2027, Colorado’s Attorney General shall adopt rules to clarify and implement these requirements. Id.

On August 11, 2026, the Colorado Department of Law released a set of proposed rules under the ADMT Act and Chatbot Safety Act that, among other things, would clarify and implement rules regarding reviewer standards, commercial reasonableness, response requirements, and documentation requirements, as discussed below.

Who Can Perform A Meaningful Human Review?

The proposed rules provide that an individual who conducts a meaningful human review must meet the following criteria:

• Independence – The reviewer must be an independent reviewer who did not make the original decision and who is not a subordinate of the original decision-maker, whenever feasible.
• Subject matter understanding – The reviewer must have a level of subject matter understanding that is commensurate with the nature of, and negative consequences resulting from, the adverse outcome of the consequential decision being reviewed.
• Sufficient training – The reviewer must be trained regarding (a) accuracy and objectivity in decision-making; (b) information considered by the AI employment tool; (c) the output’s intended use, material limitations, and categories of inputs; and (d) the subject matter at issue at a level that would enable the reviewer to identify whether review of additional available primary evidence would be valuable, and to review and understand that evidence.
• Sufficient authority – The reviewer must not be subject to steering by the upper management that would influence the reviewer’s decision, and they must be shielded from potential retaliation.
• Unassisted by AI – The reviewer may not use any ADMT in conducting his or her review.

When Is Meaningful Human Review Commercially Reasonable?

Under the ADMT Act, employers receiving requests to review adverse outcomes materially influenced by an AI employment tool must conduct a meaningful human review and reconsideration of such outcome “to the extent commercially reasonable.” 

The proposed rules provide that an employer bears the burden of demonstrating that meaningful human review is not commercially reasonable.

To meet this burden, employers must use specific evidence, and consider the following factors, weighed together, with no single factor being dispositive:

• type of review required — i.e., either (a) if the circumstances and request indicate that the AI tool may have malfunctioned, then correcting and re-performing the decision-making process; or (b) if additional evidence is submitted by the employee or job applicant, then considering whether that evidence changes the adverse outcome;
• magnitude of harm resulting from the adverse outcome;
• reversibility of the adverse outcome;
• value provided by the review of available primary evidence;
• employer size and capacity;
• marginal cost and technical feasibility of the review; and
• availability of qualified reviewers.

When the harm to an employee or job applicant resulting from an adverse outcome is a severe and irreversible denial of a basic human need, meaningful human review is presumed to be commercially reasonable. This presumption of commercial reasonableness can be rebutted by evidence showing that the review is technically or financially impossible or could not change the adverse outcome of the consequential decision.

When And How Must An Employer Respond To A Request For Review?

Within 10 days after receiving a request for review of an adverse outcome of a consequential decision meaningfully influenced by an AI employment tool, the employer must confirm receipt of the request and provide information about how the employer will process the request.

Within 45 days of receiving the request, meaningful human review must be completed, and a response must be provided to the employee or job candidate. The response must include the reviewer’s decision to confirm or override the consequential decision, the type of review conducted, the factors considered in making that decision, and the reasons the reviewer decided to confirm or override the consequential decision. Reasons provided must be specific to the evidence provided, and not a recitation of the AI tool’s general logic. 

Where possible, an adverse outcome must be stayed pending meaningful human review.

What Documentation Of Meaningful Human Reviews Must Be Retained?

When a meaningful human review is conducted, the employer must retain a record showing:

• the reviewer identity, authority, and relevant training;
• review timestamps;
• primary evidence available to the reviewer, including information provided by the employee or job candidate;
• the reviewer’s access to the AI tool’s intended use, limitations, inputs and principal factors;
• whether the reviewer approved, modified, or overrode the output; and
• a written justification for the reviewer’s decision to approve, modify, or override the output.

Implications For Companies

Colorado’s ADMT Act and Chatbot Safety Act and their corresponding proposed rules add significant operational obligations and compliance burdens for companies using automated decision-making technology to meaningfully influence consequential decisions, and using chatbots, respectively. 

This blog post identified one component of these multifaceted burdens relating to one type of ADMT – meaningful human review and consideration of employment decisions materially influenced by AI employment tools.

Although the ADMT Act and Chatbot Safety Act do not provide a private right of action, violations may be prosecuted by the Colorado Attorney General and are treated as a deceptive trade practice under the Colorado Consumer Protection Act, carrying civil penalties of up to $20,000 per violation.

Companies using or considering using ADMT or chatbots should consider whether to comment on the proposed rules before the comment period closes on October 26, 2027, should continue to monitor the content of the rules as they may evolve during the notice and comment period, and should prepare for complying with the rules by the time they are scheduled to come into effect on January 1, 2027.