Colorado’s Department of Law released interim draft rules for the ADMT Act on Oct. 6, and the quiet move inside the draft is the one every small operator should read first: Rule 6 flips “materially influences” into a rebuttable presumption. If an AI output constrains options, sets a threshold, or produces a rank, score, classification, or recommendation that a human reviewer then acts on consistently, it’s presumed to influence the decision. The burden of proof lives with the business.

That’s a structural shift from the August draft and the 100-day runway. To rebut, a deployer must document that the output was a “de minimis factor” or show that other independent factors played a “significantly larger role.” Covered domains, per Mondaq, span employment, housing, lending, insurance, healthcare, and essential government services. An owner using an off-the-shelf lead-scorer or applicant screener is a deployer.

The operational obligations are concrete. Adverse-outcome notices. Personal-data requests answered within 45 days. Human-review requests acknowledged within 10 days, completed within 45, by an independent reviewer the ADMT can’t assist. The AG declined a formal enforcement grace period but said it’ll focus on “egregious and intentional” violations, which is less reassuring than it reads.

SB 26-189 and HB 26-1263 both take effect Jan. 1, 2027. The Oct. 26 hearing at 1300 Broadway in Denver, with comments open until 11:59 p.m. MST, is the last formal input before the rules harden. xAI’s constitutional challenge remains pending in the District of Colorado, which is its own tell about where the industry thinks the fight actually goes. See also the earlier chatbot comment window.

Sources

Sources