On September 23, 2026, the Colorado Department of Law posted its revised proposed rules under the Automated Decision-Making Technology Act. The rules specify the human reviewer in more detail than they specify the system being reviewed. They describe a person who holds declared properties and the power to overturn what the system decided.
What the rule asks of the reviewer.
The proposed rules are 4 CCR 904-6, filed August 11, 2026, implementing two statutes enacted this year. The Automated Decision-Making Technology Act, SB 26-189, and the Conversational Artificial Intelligence Services Act, HB 26-1263. The formal rulemaking hearing is October 26, 2026 in Denver and by videoconference, with comments on revisions to be presented at the hearing due October 5. The ADMT Act takes effect January 1, 2027 whether or not the rulemaking has closed.
Rule 7.7 governs meaningful human review of an adverse outcome a covered system materially influenced. The reviewer must be independent of the original decision and not a subordinate of the original decision-maker, whenever feasible. The reviewer must hold competence commensurate with the nature of the adverse outcome and the consequences flowing from it. The reviewer must be trained on accuracy and objectivity in decision-making, on what the system considered, on the output's intended use and material limitations and input categories, and on the subject matter itself.
Three more requirements sit on top of that. The reviewer must not be subject to steering by upper management and must be shielded from retaliation. The reviewer must hold the authority to overturn the automated outcome. The reviewer may not use any covered automated system in conducting the review. Requests are acknowledged within 10 days and reviews completed within 45, with the adverse decision stayed pending review where the rule requires it, and the burden of showing that meaningful human review is not commercially reasonable falls on the employer.
A reviewer with properties is an object in the design.
Every one of those requirements is checkable before any particular decision arrives, and none of them is checkable from inside the request. Independence is a fact about an organizational chart. Competence is a fact about a person's history. Training is a fact about a record somebody maintains. Freedom from steering is a fact about reporting lines and retaliation policy.
That has a consequence for how a deployment is built. The party who receives a decision the system was not allowed to make has to be named before the run, in the same place and with the same standing as the bound the system is governed against. A target chosen at the moment of the decision, by whoever is on call, satisfies none of the four.
The party on the far side of a withheld action is a declared object, not a queue. Naming that party is part of declaring the envelope, and the person who declares it is answerable for the properties Rule 7.7 lists.
Deciding a review is required is not conducting the review.
The rule bars the covered system from assisting in the review. Read plainly, it prevents the thing that produced the outcome from also producing the verdict on whether the outcome was right.
A runtime governor never decides the case. It sets how much authority the next action carries and whether it carries any at all. When it grants nothing, it has established that the action is not the system's to take, and that establishes the condition Rule 7.7 attaches to rather than discharging it. Deciding that a decision requires a person is a different act from making it.
What the governor assessed travels with the withheld action as evidence of why it was held. It does not stand in for the determination the rule asks a person to reach. Wayfinder sits above control and below intelligence, and the reviewer sits above Wayfinder.
The step nobody specified.
The rule attaches to an adverse outcome and to a consumer who asks for reconsideration. Before either of those exists, something already decided that this particular decision would be made by the system rather than by a person. In most deployments that selection is a threshold in a configuration file, edited by whoever owns the deployment, on no schedule.
That selection is itself an automated decision, and it materially influences a consequential decision. It runs earlier than every obligation the rule names and it is the only step in the sequence that no part of the rule reaches.
A governor makes that selection per action, in advance, from inside the decision path. The action cannot start until the grant is set, so the routing is decided rather than discovered. Colorado's clock for the reviewer is measured in days because it describes a consumer appeal. The clock for the selection is the interval between one action and the next.
Approve, modify, or overturn.
The rule gives the reviewer authority to overturn the automated outcome. Reviewers who matter in practice hold a third power, which is to let the decision stand on narrower terms than the system proposed.
That is the same three-valued answer a governor returns on every action. Grant nothing. Grant in full. Grant a bounded amount. A reviewer whose only two options are approve and overturn will approve matters that should have been narrowed, because narrowing is not on the form.
A grant a person sets is the same kind of object the governor sets, and it enters the run the same way. As a ceiling on the next action, not as a note filed alongside it.
What we are building.
Wayfinder Systems Group builds a runtime governor. It observes a system's behavior against an envelope declared before the run, sets how much authority the next action carries, and enforces that bound in front of the actuator rather than beside the log. A governor can grant less authority on a decision whose trust reads higher, because its forward look caught a divergence coming before the outcome landed. The same forward look decides when the next action is not the system's to take, and the party that receives it was named before the run. We call her Velma. The signed record is the evidence of what was granted and to whom.
Thirty minutes. Architecture, not sales.
A conversation about where the routing decision is made in your deployment today, who is named to receive a withheld action, and whether that party's properties are declared anywhere a regulator could read them.
JonathanLuethke@WayfinderSystemsGroup.com
