Colorado's AI Accountability Regime

Overview

Colorado's SB26-189 and Chatbot Safety Act, read with proposed 4 CCR 904-6 details, turn AI governance into an observable chain: test coverage, preserve handoffs and versions, trace decisions, explain adverse outcomes, enable review, and test chatbot safeguards. The series moves from legal boundaries to consumer remedies and product controls, while preserving that rulemaking is not all final law and records do not prove fairness or safety.

Learning outcomes

Coverage turns on an output's material influence on a consequential decision, not on an AI label.
Accountability depends on records linking intended use, versions, inputs, actors, decisions, and later review.
Specific explanation, factual correction, and Meaningful Human Review make consumer remedies usable, but do not guarantee reversal.
Proposed chatbot controls pair bounded-product analysis with age assurance, disclosure, tested safeguards, and crisis response that preserves uncertainty.

Keywords

ColoradoSB26-189Chatbot Safety ActADMTAccountabilityTraceabilityTransparencyMeaningful Human ReviewAge assuranceCrisis protocols

Chapters

Colorado's AI Law in Practice

How two laws become observable duties, records, and safeguards

9 min · How two laws become observable duties, records, and safeguards

9 min

Colorado's ADMT Coverage Test

When an output materially influences a consequential decision

10 min · When an output materially influences a consequential decision

10 min

Colorado AI Decision Records

Preserving developer-to-deployer evidence across versions

9 min · Preserving developer-to-deployer evidence across versions

9 min

Colorado ADMT Housing Deposit

Tracing one scored decision from applicant data to remedy

8 min · Tracing one scored decision from applicant data to remedy

8 min

Colorado ADMT Consumer Remedies

From advance notice to correction and reconsideration

9 min · From advance notice to correction and reconsideration

9 min

Colorado Meaningful Human Review

When approval can actually change an adverse outcome

9 min · When approval can actually change an adverse outcome

9 min

Colorado Chatbot Safety Boundaries

Age assurance, disclosure, minor safeguards, and crisis response

12 min · Age assurance, disclosure, minor safeguards, and crisis response

12 min

Colorado AI Accountability Limits

What records, review, and testing can—and cannot—prove

11 min · What records, review, and testing can—and cannot—prove

11 min

Preview every chapter · Unlock full text, audio, and video from chapter 4

Plans

Ch.1-3 FreeCh.4-8 Premium

FAQ

When does Colorado's AI law cover an automated decision?

Colorado's ADMT law covers a use when personal data produces a computational output that materially influences a consequential decision in a covered domain, absent an exclusion.

What can a Colorado consumer request after an adverse ADMT outcome?

A Colorado consumer can request system and input information, correct factually inaccurate personal data, and seek Meaningful Human Review through the deployer's process. The statute requires post-outcome disclosure within thirty days.

Does human approval make an AI decision compliant in Colorado?

No. Human approval is not Meaningful Human Review unless a trained, sufficiently independent reviewer can examine relevant evidence, understand limitations, and change the outcome; it does not replace other legal duties.

How do Colorado's chatbot rules address safety?

The proposed rules address chatbot safety through product-boundary analysis, age assurance, recurring disclosure, minor safeguards, crisis protocols, testing, and reporting; they do not prove every system is safe.