TOP
Big Jerm Says.

Most AI governance content is commentary about rules. This one is different: it’s a look inside an actual filing — the public comment Combs Intelligence Group LLC submitted to the Colorado Attorney General on the proposed rules implementing Colorado’s Automated Decision-Making Technology Act.

The comment was filed Saturday, October 24, 2026 — two days ahead of the deadline — and it’s part of the public record under CIG’s name. Here’s what the rules would do, what we argued, and why it matters if your business uses AI anywhere near a consequential decision.

What Colorado is building

The proposed rules (4 CCR 904-6) implement two laws: the ADMT Act (SB26-189) and the Chatbot Safety Act (HB26-1263). The core mechanism: when an automated system’s output “materially influences” a consequential decision about a person — think employment, housing, credit, healthcare — duties attach to the business deploying it.

The provision that does the most work is the Rule 6.1 material-influence presumption. Under Rule 6.1A, if a human reviewed an AI output and the outcome is consistent with that output, the output is presumed to have materially influenced the decision. Rule 6.1B lists evidence a business can use to rebut that presumption.

The rules carry a January 1, 2027 effective date, with enforcement currently stayed in xAI v. Weiser.

The four positions CIG filed

1. Judge the decision record, not the review.

The presumption as drafted treats human review itself as the trigger: route an output past a qualified reviewer, and you’re presumed influenced — in the same position as a business that rubber-stamps everything. The AG’s own worked example makes the point vividly: a manager who reviewed all the underlying files and interviewed every candidate is still presumed influenced. A standard that penalizes oversight will produce less of it. The test should be what the contemporaneous decision record shows about the output’s actual role.

2. The rebuttal path needs a sufficiency standard.

Rule 6.1B helpfully lists rebuttal evidence — the decision-maker didn’t consult the output, a hard cut-off was consistently applied, an independent review of substantial information happened, and so on. But it never says how much evidence is enough, or what makes evidence “persuasive” under Rule 6.1B.5. We asked for three fixes: a stated sufficiency standard; confirmation that partial satisfaction of Rule 6.1B.3 can still rebut; and worked examples of successful rebuttals in Rule 6.2, matching the presumption examples already there. Without them, a business can’t know in advance whether its decision record will hold — and in the cases that matter most, the presumption operates as a conclusion.

3. A safe harbor for documented human review.

Businesses that maintain a named reviewer with genuine override authority, override logging, and a per-decision record should receive a defined compliance benefit in exchange for that real oversight. Certainty for documented review is the mechanism most likely to produce meaningful human review in practice, rather than review theater.

4. A supplemental comment window if the date moves.

With enforcement stayed and the effective date in question, we asked the AG to confirm that any shift in the January 1, 2027 date comes with a supplemental comment window on revised text.

How the filing was built (and the mistake we caught)

One process note, because it’s the part most relevant to anyone doing this work: the obvious full-text PDF of these rules circulating online is the August proposal. It contains no material-influence presumption at all. The presumption lives in the Attorney General’s October 6 interim redline. We caught the wrong document before drafting against it, corroborated the presumption’s structure through law-firm reporting, then read the interim redline directly — and found that Rule 6.1B’s rebuttal list was more developed than our early draft assumed, which changed position 2 from “create a rebuttal path” to the sharper “give the existing path a sufficiency standard.”

That’s the whole discipline in one story: argue from the operative text, or don’t argue. It’s the same standard behind the practical governance system in AI Governance for Small Business — and behind the wider system this site is built on, laid out in the pillar piece here: https://bigjermsays.com/how-to-make-money-with-ai-build-a-system-not-a-side-hustle/

What this means for your business

If your company uses AI output in decisions about people, the Colorado model — decision records, named reviewers, override logs — is the template other states are watching. You don’t need a legal department to start; you need a register of where AI is used, an approval rule for consequential uses, and a record habit. That’s buildable this quarter, at small-business scale.

Start here

The system CIG uses with clients — policy, register, approval workflow, and the templates to run it — is packaged here:

The AI Governance Starter Kit, $197

It won’t file your comments for you. It will make sure that when a rule like Colorado’s asks “show me your decision record,” you have one.

This article describes a public filing and proposed rules; it is not legal advice. Combs Intelligence Group LLC’s comment is part of the public record in the Colorado ADMT and Chatbot Safety rulemaking (4 CCR 904-6).

FAQ

When do the Colorado ADMT rules take effect?

January 1, 2027, as proposed — though enforcement is currently stayed in xAI v. Weiser, which is why CIG’s comment asked for a supplemental comment window if the date moves.

What counts as “materially influence”?

Under the proposed Rule 6.1A, an AI output is presumed to materially influence a consequential decision when a reviewed output is consistent with the outcome. Rule 6.1B lists evidence that can rebut the presumption. The statutory definition sits at C.R.S. § 6-1-1701(13).

Does this apply to small businesses?

The Act’s duties center on deployers of ADMT in consequential decisions. The practical takeaway for a small firm is the same regardless of where the thresholds land: keep a register, name a reviewer, log overrides. Those habits are cheap now and expensive to retrofit.

Can I read the comment itself?

Yes — comments in this rulemaking are posted publicly under the commenter’s name and organization. Search the Colorado AG’s ADMT rulemaking page for Combs Intelligence Group LLC, filed October 24, 2026.

Share Article:
Big Jerm

Jeremy Combs (Big Jerm) is the founder of Combs Intelligence Group LLC (combsintelligencegroup.com), a fractional CTO and AI governance practice that helps nonprofits, tribal enterprises, and founder led businesses adopt AI with real controls in place. He is also a co-founder of G1FTI, a crowdfunding platform for verified world changing missions. His background includes years of HIPAA certified technology work supporting tribal healthcare, and he is certified to teach the Anthropic AI Fluency framework. He holds an Associate's in Software Product Management from Oklahoma State University, Spears School of Business. Big Jerm Says is where he shares what he is building and learning in public.

© 2026 Big Jerm Says. All rights reserved.

More than opinions. Real talk.

GitHub