Quick Hits
- The Colorado Attorney General’s Office will hold a rulemaking hearing on October 26, 2026, to gather input on new regulations for AI-driven employment tools before they take effect on January 1, 2027.
- The attorney general’s office is seeking input on how to define when a covered AI tool “materially influence[s]” a decision, thereby triggering obligations under the state’s new AI law, how to allocate responsibilities when deployers rely on tools operated by third-party vendors, and whether the rules should detail requirements for developer disclosures about known limitations of AI tools.
On August 11, 2026, the Colorado Attorney General’s Office Department filed proposed rules, titled “Automated Decision-Making Technology and Conversational Artificial Intelligence Service Rules.” The proposed rules seek to implement and clarify the compliance obligations under Colorado Senate Bill (SB) 26-189, the Automated Decision-Making Technology in Consequential Decisions (ADMT) Act, and House Bill (HB) 26-1263, the Chatbot Safety Act, both of which were signed into law by Governor Jared Polis in May 2026 and take effect on January 1, 2027.
According to the hearing notice, the attorney general’s office is seeking input about specific issues that were not fleshed out in the proposed rules that could have a particular impact on compliance obligations for employers. Specifically, the office is seeking input on the standard for “materially influence” in a consequential decision, the threshold that triggers obligations under the ADMT Act, multiparty arrangements with automated decision-making technology (ADMT) vendors, which are not explicitly contemplated by the act despite being typical in the employment context, and the amount of detail that developers should be required to provide to employers about known risks of AI tools.
‘Materially Influence’ Standard
The ADMT Act repealed and replaced the state’s landmark 2024 AI law, and will impose a new multistage notice framework and rights for affected “consumers,” including employees and job applicants, to seek information about the tool and “meaningful human review.” The act’s obligations are triggered when an ADMT “is used to materially influence a consequential decision” such as a hiring, promotion, or termination decision. The act defines “materially influence” to mean the ADMT output is “a non-de minimis factor” affecting the outcome of the consequential decision, “including by constraining, ranking, scoring, recommending, classifying, or otherwise meaningfully altering how a consequential decision is made.”
However, the attorney general’s office is seeking additional input on how to clarify this “materially influence” standard, specifically what would constitute “a non-de minimis factor.” The attorney general’s office is considering two competing standards and specifically asking which one to adopt.
Standard 1: The Lower Threshold
Under this standard, a “de minimis factor” is one that has only a “trifling, trivial, or incidental” impact on the outcome. Critically, a factor would not be deemed de minimis simply because other factors played a more significant role. The standard would create a rebuttable presumption of material influence whenever the ADMT output pertains to the individual, is reviewed by (or screens data for) the decision-maker, and is consistent with the outcome.
Employers would be able to rebut the presumption with evidence that the decision-maker recorded an independent judgment before seeing the ADMT output, did not view the output, made a decision inconsistent with the output, or relied on other previously documented and consistently applied dispositive factors.
Standard 2: The Higher Threshold
Under this standard, a “de minimis factor” would be one that is “not a substantial factor” in the consequential decision. A factor could be de minimis if other relevant factors independent of the ADMT output played a “significantly larger role.” The presumption of material influence uses the same trigger—that the output pertains to the individual, is reviewed by, or screens for the decision-maker, and is consistent with the outcome.
But the rebuttal standard is notably different. Under Standard 2, rebuttal requires showing that the decision-maker conducted an independent review of “substantial information” other than the ADMT output, that the outcome was consistent with that other information, that the decision-maker had authority to decide based on independent review, and that the decision-maker had the subject matter understanding necessary to review and understand the substantial relevant information.
Under the act, “materially influence” does include “incidental, trivial, or clerical uses.” As a practical matter, however, adoption of the lower threshold seemingly would bring a large percentage of use cases of ADMTs in the employment context within scope of the act.
Multiparty Arrangements and ADMT Vendors
The ADMT Act does not explicitly contemplate scenarios in which deployers (i.e., employers) do not directly operate the ADMT that materially influences their decisions, yet some employers rely upon ADMT tools operated by third-party vendors. The attorney general is asking:
- Should rules clarify that a deployer may run an ADMT through a vendor?
- Should rules mandate that deployers remain responsible for fulfilling all deployer obligations regardless of vendor involvement?
- Should ADMT vendors be required to assist deployers in responding to consumer requests for personal data and correction?
- In the staffing agency scenario—where the agency operates the ADMT and the employer relies on its output—which party bears deployer obligations?
- Should vendor obligations be modeled on processor obligations under the Colorado Privacy Act?
These questions are crucial to regulating the procurement and use of AI in the employment context. The allocation of compliance responsibility between the employer-deployer and the vendor will be a major factor in cost, contracting, and risk.
ADMT Developer Obligations
The ADMT Act requires developers to make certain information available to deployers (i.e., employers), but the attorney general is seeking input about whether the rules should incorporate greater specificity about the content of these disclosures. With respect to this issue, the attorney general is seeking input about the following topics:
- Should rules require a developer to disclose any testing performed, how the testing was performed and the results of such testing?
- With what level of specificity should developers be required to disclose the categories of data used to train the ADMT?
- Should the rules specify the form for developers to provide information or the channels through which such information is provided, if rules are adopted to require more detail about known limitations on the ADMT?
Because employers rely upon information provided by developers in making procurement decisions, the adoption of rules imposing more robust disclosure requirements on developers doing business in Colorado could aid employers in evaluating which tools to purchase.
Next Steps
The public hearing is scheduled for October 26, 2026, at 10:00 a.m. According to the hearing notice, comments must be submitted by September 4, 2026, to be considered in a second proposed draft to be circulated by September 23, 2026, and discussed at the hearing. However, any written comments received on or before 11:59 p.m. on October 26, 2026, will be considered in the final set of rules. If the rulemaking hearing continues beyond October 26, 2026, the attorney general will consider any written comments received before 11:59 p.m. MT on the last day of the formal rulemaking hearing.
The Colorado ADMT proposed rules are the latest in a growing trend of state regulations impacting employers’ use of AI in the workplace to make, facilitate, or influence hiring, promotion, and termination decisions about individual employees or job applicants. Although some modifications to the initial draft may be made, these proposed rules provide some indication of the requirements that will apply to employers doing business in Colorado that use ADMTs to evaluate job applicants or employees residing in Colorado.
Employers in Colorado and other stakeholders may want to consider attending and/or testifying at the hearing and consider submitting formal written comments on the proposed rules.
Ogletree Deakins’ Artificial Intelligence and Innovation Practice Group will continue to monitor developments and will provide updates on the Artificial Intelligence and Innovation, Colorado, Cybersecurity and Privacy, and Employment Law blogs as additional information becomes available.
Follow and Subscribe
LinkedIn | Instagram | Webinars | Podcasts
