Colorado SB 26-189 · ADMT in consequential decisions
Correction and meaningful human review under SB 26-189
Under C.R.S. 6-1-1705(1)(a), a consumer who experiences an adverse outcome from a consequential decision materially influenced by a covered ADMT may request, and the deployer shall provide, instructions for requesting and correcting personal data and an opportunity for meaningful human review and reconsideration, the latter to the extent commercially reasonable. The statute does not require correction of opinions, predictions, scores, or protected evaluations. The Attorney General's rules on these rights are still a proposed interim draft.
Last reviewed: November 13, 2026 · Published
Not legal advice. This page is general information about Colorado SB 26-189 for engineering, product, and governance teams. It is not legal advice and does not create an attorney-client relationship. CertifiedData supports evidence for record keeping and disclosures; it does not determine whether a system is covered by the law and does not make any organization compliant. Confirm your obligations with qualified Colorado counsel.
Attorney General rule text referenced on this page is a proposed interim draft (released October 6, 2026) for 4 CCR 904-6 and is subject to change. As of November 13, 2026 no adopted final rule was shown on the Attorney General's rulemaking page.
What 6-1-1705 gives the consumer
The right is triggered when a consumer experiences an adverse outcome resulting from a consequential decision in which a covered ADMT materially influenced the decision (C.R.S. 6-1-1705(1)(a)). On request, the deployer shall provide two things.
- 6-1-1705(1)(a)(I): instructions for requesting personal data and for correcting factually incorrect or materially inaccurate personal data used in a consequential decision that used a covered ADMT, consistent with C.R.S. 6-1-1306.
- 6-1-1705(1)(a)(II): an opportunity for meaningful human review and reconsideration of the consequential decision, to the extent commercially reasonable.
Note the wording of (1)(a)(I): the deployer provides instructions for requesting and correcting personal data. The statute frames the obligation around instructions and the existing 6-1-1306 mechanism, not a standalone new correction procedure.
The adverse-outcome disclosure required by 6-1-1704(3)(c) must include an explanation of these consumer rights and how to exercise them, so the notice and the rights process need to be built together.
Exceptions, limits, and the FERPA path
Exceptions that do not apply to correction
For purposes of 6-1-1705(1), the exceptions to the definition of consumer in C.R.S. 6-1-1303(6)(b) and the exceptions in 6-1-1304(2)(k), (2)(n), and (2)(o) do not apply to the right to request correction of factually incorrect or materially inaccurate personal data (6-1-1705(1)(b)). In practice, a deployer should not assume that a Colorado Privacy Act exemption removes the correction right here. Confirm the specifics with counsel.
What correction does not cover
Subsection (1)(a) does not require correction of opinions, predictions, scores, or protected evaluations (6-1-1705(1)(c)). A model score can be the thing a consumer disputes, but the right is to correct the underlying personal data, not to edit the score itself. Reconsideration is the separate route that can address the outcome.
FERPA path for education decisions
For a consequential decision relating to education, a deployer subject to FERPA complies with 6-1-1705(1) through its existing student record inspection, review, and amendment procedures and any applicable district complaint or appeal process, if it offers a reasonable mechanism for a parent, guardian, or eligible student to request correction of materially inaccurate personal data and reconsideration where applicable (6-1-1705(2)(a)). It is not required to build a separate or duplicative process if it has already established one to comply with FERPA (6-1-1705(2)(b)).
What counts as meaningful human review
C.R.S. 6-1-1701(15) defines meaningful human review as review by an individual designated by the deployer who has authority to approve, modify, or override a consequential decision and who meets four conditions.
| Statute | Requirement |
|---|---|
| 6-1-1701(15)(a) | Considers relevant, available primary evidence. |
| 6-1-1701(15)(b) | Is trained to conduct the review. |
| 6-1-1701(15)(c) | Does not default to the system output. |
| 6-1-1701(15)(d) | Has access to sufficient information to understand the output's intended use, material limitations, and categories of inputs, and the principal factors used to generate the output, without requiring disclosure of proprietary source code, model weights, or other trade secrets. |
The deployer therefore needs a named reviewer role with real authority over the decision, a training record, and access to the information in (d). Developers are directed to supply some of that information: C.R.S. 6-1-1702(1)(d) requires developer documentation to include instructions for the deployer's appropriate use, monitoring, and meaningful human review, where applicable. Developer documentation is the natural source for what a reviewer needs under (15)(d).
The commercially reasonable qualifier
The statute attaches the phrase to the human-review opportunity only: "to the extent commercially reasonable" (6-1-1705(1)(a)(II)). The statute does not define the phrase. It does not say the qualifier applies to the data request and correction instructions in (1)(a)(I). Whether a particular deployer may decline or limit review on this ground is a fact-specific question for counsel, and the Attorney General's proposed rules address it (see below).
Proposed AG draft Rule 8 (subject to change)
Proposed interim draft (October 6, 2026), subject to change. Not adopted law. As of November 13, 2026 no adopted final rule was shown on the Attorney General's page. The draft is a redline containing tracked insertions and deletions, so specific figures and wording may differ in the final rule. Verify against the current Attorney General text.
Rule 8 of the interim draft of 4 CCR 904-6 is titled ADMT Consumer Rights. It states that its authority is C.R.S. 6-1-1705(3) and that its purpose is to clarify and implement 6-1-1705. In summary, the draft covers the following topics.
- Submitting requests: the adverse-outcome disclosure must lead the consumer to a request method. The draft describes multiple request methods that reflect how consumers normally interact with the deployer, monitoring of those channels by someone able to process requests, and a low-friction process. It would bar requiring a new user account, though an existing password-protected account may be required.
- Right to request personal data: the draft describes instructions and response requirements, including that specific pieces of personal data include final rank, score, classification, recommendation, prediction, profiles, or other inferences and the ADMT inputs about the individual, delivered in an intelligible format. Where the deployer does not hold the data, it would point the consumer to third parties that do.
- Correction: the draft describes handling of correction requests, documentation a deployer may ask for, and denial where the deployer reasonably determines the data is more likely than not accurate, with the denial and its reasons documented. It restates that opinions, predictions, scores, and protected evaluations need not be corrected.
- Authentication: a commercially reasonable method, a bar on charging fees to authenticate, limits on using authentication data for other purposes, and a process when authentication fails.
- Meaningful human review and reconsideration: requests through the same method as other rights; reconsideration based on corrected data; reviewer standards (actual authority, training, subject-matter understanding, protection from management steering); a statement that ADMT may not assist the review, with an exception for tools that do not meet the ADMT definition; types of review; and a response stating the decision, type of review, factors considered, and reasons.
- Commercial reasonableness: the draft lists factors weighed together, presumes review is commercially reasonable when the harm is a severe and irreversible denial of a basic human need, and places the burden on the deployer to show with specific evidence that review is not commercially reasonable.
- Documentation: when a review is conducted, the draft would require the deployer to retain a record showing reviewer identity, authority, and relevant training; review timestamps; primary evidence available to the reviewer; the reviewer's access to the ADMT's intended use, limitations, inputs, and principal factors; whether the reviewer approved, modified, or overrode the decision; and a written justification.
- FERPA: the draft restates the education path in 6-1-1705(2).
The draft also contains response timing for acknowledging and completing requests. Because the redline mixes proposed insertions and deletions, confirm those periods in the Attorney General's current text before relying on them.
How CertifiedData supports evidence
CertifiedData helps you produce records for correction and review requests; it does not run your review or decide whether a review was adequate. Decision Ledger records are canonicalized (RFC 8785 JCS), SHA-256 hashed, Ed25519 signed, and hash-chained, so modification of a recorded decision is detectable. The audit bundle export is a ZIP with the certificate, transparency log entry, public key, and verification instructions, and the public verifier at /verify lets a third party check a record independently.
The Colorado ADMT evidence pack is an authenticated export that packages Colorado-specific decision metadata, including the ADMT name, version, and developer, adverse-outcome flag, disclosure timestamp and 30-day deadline check, data categories, trade-secret flag, and consumer correction and human-review requests, together with hash-chain and signature verification material. It records what the deployer supplies. Documentation is at /compliance/colorado-sb-26-189/evidence-pack.
Frequently asked questions
Does SB 26-189 require a deployer to change a model's score if a consumer disputes it?
No. C.R.S. 6-1-1705(1)(c) says the correction right does not require correction of opinions, predictions, scores, or protected evaluations. The right concerns factually incorrect or materially inaccurate personal data, and the consumer may separately seek meaningful human review and reconsideration.
Who can perform meaningful human review?
An individual designated by the deployer who has authority to approve, modify, or override the decision, considers primary evidence, is trained, and does not default to the system output (C.R.S. 6-1-1701(15)). The AG's proposed draft would add further reviewer standards, which are subject to change.
Does the human review right apply in every case?
The statute grants an opportunity for meaningful human review and reconsideration only "to the extent commercially reasonable" (6-1-1705(1)(a)(II)). The AG's interim draft proposes factors and a presumption for severe harms, but that text is not adopted.
Is the AG's Rule 8 final?
No. It is a proposed interim draft released October 6, 2026. As of November 13, 2026 no adopted final rule was shown, and the draft states a proposed effective date of January 1, 2027.
Sources
- SB 26-189 enacted session law (S.L. 2026, ch. 131; C.R.S. 6-1-1701 to 6-1-1709), approved May 14, 2026
- SB 26-189 enrolled text (LegiScan)
- SB 26-189 fiscal note (Senate 2nd reading, May 6, 2026)
- Colorado Attorney General — ADMT and chatbot rulemaking page
- AG proposed rules, interim draft redline (Oct 6, 2026) — 4 CCR 904-6, proposed
- AG Regulatory Analysis (Oct 21, 2026)
- Colorado Secretary of State eDocket, tracking no. 2026-00360