Ethics and Compliance

Colorado's AI Law Rewrite: What Law Firms Should Know

Colorado replaced its 2024 AI Act with SB26-189. Legal services left the covered list, hiring tools did not, and duties start January 1, 2027.

AssistCall research teamPublished Reviewed by Faruque Amin, VP of Sales and Marketing9 min read

On this page
  1. The law your memo describes never took effect
  2. What changed between the two versions
  3. Where a law firm is still a deployer
  4. Phones, intake and scheduling
  5. What to do before January 1, 2027
  6. Enforcement, cure periods and contracts
  7. Frequently asked questions
  8. References
Ethics and Compliance: Colorado's AI Law Rewrite: What Law Firms Should KnowJan 1, 2027: The day Colorado's rewritten AI law starts applying to consequential decisionsETHICS AND COMPLIANCEColorado's AI Law Rewrite:What Law Firms Should KnowJan 1, 2027The day Colorado's rewritten AI lawstarts applying to consequentialdecisionsAva for LawyersInsightsEthics and Compliance: Colorado's AI Law Rewrite: What Law Firms Should KnowJan 1, 2027: The day Colorado's rewritten AI law starts applying to consequential decisionsETHICS AND COMPLIANCEColorado's AI LawRewrite: What LawFirms Should KnowJan 1, 2027The day Colorado's rewritten AI lawstarts applying to consequentialdecisionsAva for Lawyers
Figure source: Colorado General Assembly, 2026[1]

Key takeaways

  • SB26-189 repealed and reenacted Colorado's 2024 AI Act before it ever took effect; the new law applies to consequential decisions made on or after January 1, 2027[1].
  • Legal services were one of eight covered areas in the 2024 Act[2] and are not among the seven covered domains in the rewrite[1].
  • A firm can still be a deployer as an employer, because employees and Colorado-resident job applicants count as consumers[1].
  • Routine scheduling, administrative routing and customer service triage are carved out of consequential decisions[1], which is where most phone and intake automation sits.
  • The 2024 Act's relief for deployers under 50 full-time employees[2] is not in the new text; coverage turns on what a tool does, not on firm size.

The law your memo describes never took effect

If your firm wrote an AI compliance memo in 2024 or 2025, it describes a statute that no longer exists. Colorado's SB 24-205 was approved on May 17, 2024, with most duties set to begin on February 1, 2026[2]. A 2025 special-session bill pushed that start to June 30, 2026[3]. Before that date arrived, the legislature passed SB26-189, which the Governor signed on May 14, 2026. It repealed the 2024 provisions and reenacted that part of the statute as a law on automated decision-making technology, or ADMT[4].

The new version takes effect on January 1, 2027 and applies to consequential decisions made on or after that day[1]. A few provisions took effect on passage, chiefly the Attorney General's rule-making: rules on post-adverse-outcome disclosures and on consumer rights are due on or before January 1, 2027[1]. The working window for a firm is short, and some of the detail will only arrive with those rules late in 2026.

From the 2024 AI Act to SB26-189
  1. May 17, 2024

    SB 24-205 approved, with most duties set to start February 1, 2026[2]

  2. Aug 28, 2025

    SB 25B-004 approved, moving the start date to June 30, 2026[3]

  3. May 14, 2026

    SB26-189 signed, repealing and reenacting the law as an ADMT statute[4]

  4. Jan 1, 2027

    The new law applies to consequential decisions made from this date[1]

Dates from the Colorado General Assembly's bill records and signed acts.

Two things follow. Anything in your AI use policy, engagement letters or vendor contracts that cites SB 24-205 needs a fresh look. And advice given to clients under the old framework, about impact assessments, risk-management programs or the small-deployer exemption, should be revisited before anyone relies on it again.

What changed between the two versions

The rewrite is narrower and more procedural. The 2024 Act built a duty of reasonable care around high-risk AI systems, with risk-management programs, impact assessments and annual reviews for deployers[2]. SB26-189 drops that architecture. It asks deployers to give notice, explain adverse outcomes, offer correction and human review, and keep records[1].

Summary of the signed texts of SB 24-205 and SB26-189. Read the statute itself before advising a client.
IssueSB 24-205 (2024, never in force)SB26-189 (applies from Jan 1, 2027)
What it regulatesHigh-risk AI systems that make, or are a substantial factor in, consequential decisionsADMT used to materially influence a consequential decision
Covered areasEight, ending with “a legal service”Seven: education, employment, residential housing, financial or lending services, insurance, health care, essential government services
Telling people they are dealing with AIRequired for any consumer-facing AI system unless obviousNo stand-alone duty; notice is tied to covered decisions
Small-deployer reliefSome duties waived under 50 full-time equivalent employeesNo headcount exemption in the text
After a bad outcomeNotice, correction and an appeal with human review if feasiblePlain-language explanation within 30 days, correction, meaningful human review
RecordsImpact assessments and annual reviewsCompliance records kept at least three years
EnforcementAttorney General; deceptive trade practiceAttorney General only; cure period until 2030; no new private right of action

The biggest change for this profession sits in one line. The 2024 list of consequential decisions ended with “a legal service”[2]. The 2026 list of covered domains has seven entries, and legal services is not among them[1]. Whatever a firm's intake technology does, the decision to take or decline a matter is no longer a covered decision under this statute.

The general disclosure rule went too. The 2024 Act would have required anyone deploying a consumer-facing AI system to tell each person they were dealing with AI, unless that was obvious[2]. The reenacted part has no stand-alone equivalent; its notice duty attaches only to covered ADMT used in consequential decisions[1]. That does not settle the question for your phone line. Utah requires a business using generative AI to say so when a person asks[5], and Florida's ethics opinion says chatbots that talk to clients or third parties must disclose that they are an AI program and not a lawyer or employee of the firm[6]. The state-by-state picture is in which states require AI disclosure, and greeting wording is in telling callers they are speaking with AI.

Where a law firm is still a deployer

A firm becomes a deployer when it uses a covered ADMT: technology whose output is a more-than-trivial factor that actually shapes a consequential decision, by constraining, ranking, scoring, recommending or classifying[1]. For most firms the live exposure is hiring. The law treats employees and Colorado-resident job applicants as consumers[1], and employment is a covered domain. A résumé screener that ranks associate applicants, software that scores staff for promotion, or a tool that flags who to let go in a slow year can each bring a firm inside the statute.

Size does not get you out. The 2024 Act waived its heavier duties for deployers with fewer than fifty full-time equivalent employees that met other conditions[2]. The reenacted text has no headcount test. That matters in a state where most law offices are small: the Census Bureau counted 3,431 offices of lawyers in Colorado in 2023, and 2,643 of them had fewer than five employees[7].

Colorado law offices by number of employees, 2023Establishments, not firms. Size classes of 250 or more employees are not disclosed.
Colorado law offices by number of employees, 2023Column chart. Under 5: 2,643; 5 to 9: 415; 10 to 19: 213; 20 to 49: 110; 50 to 99: 34; 100 to 249: 13.01,0002,0003,000Under 5: 2,6432,643Under 55 to 9: 4154155 to 910 to 19: 21321310 to 1920 to 49: 11011020 to 4950 to 99: 343450 to 99100 to 249: 1313100 to 249Colorado law offices by number of employees, 2023Column chart. Under 5: 2,643; 5 to 9: 415; 10 to 19: 213; 20 to 49: 110; 50 to 99: 34; 100 to 249: 13.Under 5Under 5: 2,6432,6435 to 95 to 9: 41541510 to 1910 to 19: 21321320 to 4920 to 49: 11011050 to 9950 to 99: 3434100 to 249100 to 249: 1313

Source: U.S. Census Bureau, 2023[7]

Show the data
EmployeesOffices of lawyers
Under 52,643
5 to 9415
10 to 19213
20 to 49110
50 to 9934
100 to 24913

The second exposure is advisory. Clients that lend, insure, lease homes, deliver health care or employ people in Colorado may be deployers, and some build or license tools, which makes them developers with documentation duties from January 1, 2027[1]. Their questions will reach employment, consumer-finance and health-care lawyers first, and a firm advising on the statute should have run its own inventory.

If a covered tool is in use, the deployer's duties are short to list and fiddly to run:

  • Before the tool is used in the decision, give a clear and conspicuous notice with instructions for getting more information; a prominent public notice near the point of interaction satisfies this[1].
  • Within 30 days of an adverse outcome, give a plain-language description of the decision and the tool's role, how to ask for details such as its name, version, developer and data categories, and an explanation of the person's rights[1].
  • On request, explain how to obtain and correct inaccurate personal data, and offer meaningful human review and reconsideration to the extent commercially reasonable[1].
  • Keep records showing compliance for at least three years after each consequential decision[1].

Meaningful human review has a real definition. The reviewer must have authority to approve, modify or override the decision, be trained to do the review, consider the relevant primary evidence, understand the tool's intended use, limitations and inputs, and not default to the system's output[1]. A partner who glances at a ranked shortlist and initials it does not meet that standard.

Phones, intake and scheduling

Most automation in a law office never comes near a consequential decision, and the statute says so. Low-stakes and routine processes are excluded, including routine scheduling, administrative routing, customer service triage, communication of decisions and workflow management[1]. The ADMT definition also leaves out a tool used solely to summarize, organize, draft, route or present information for human review, and conversational technology that answers questions or makes referrals, provided it is not configured or intended for consequential decisions and an acceptable use policy bars that use[1].

Set those carve-outs beside an ordinary intake line. An SI Employee that answers a forwarded line after hours, asks the questions the firm approved for the practice area, takes the names the firm needs for its conflict check, books a consultation and sends a summary to a person is doing scheduling, routing and triage for human review. The decision to accept the matter stays with a lawyer, and legal services are not a covered domain in any case[1]. The same holds for the night-call rules in the criminal defense firm example, where written custody criteria decide which calls are put through and everything else becomes a message.

Example: sorting a firm's own tools against SB26-189
  1. 01Your team

    List every AI tool

    Hiring, billing, intake, research, marketing: who uses it and what it outputs.

  2. 02Your team

    Covered domain?

    Employment, housing, lending, insurance, health care, education or public benefits.

  3. 03Your team

    Does it shape the outcome?

    A score, rank or recommendation that is more than an incidental factor.

  4. 04Your team

    Check the carve-outs

    Scheduling, routing, triage and summaries for human review sit outside the law.

  5. 05Your team

    Covered: build the file

    Notice, 30-day explanation, correction, human review and three-year records.

Example triage for a firm's own inventory. It is a starting point for a lawyer's analysis, not a substitute for it.

Two cautions. Write the acceptable-use limit into your AI policy so the carve-out's condition is met on paper, not assumed; writing a law firm AI use policy covers the structure. And keep the professional rules in view, because they apply whatever the statute says. California's State Bar lists autonomously facilitating client intake as an example of agentic AI and says lawyer supervision must match the system's access and autonomy[8]. To see what a lawyer actually reviews after an intake call, the summary, recording and transcript, look at what the SI Employee does on a call.

What to do before January 1, 2027

  1. Inventory tools that touch people decisions

    Start with recruiting and HR software. Ask each vendor whether its product is marketed or configured to rank, score or screen candidates or staff. If it is, treat it as likely covered when used on Colorado applicants or employees.

  2. Ask vendors for developer documentation

    From January 1, 2027 developers must give deployers a statement of intended and harmful uses, training-data categories, known limitations and instructions for use and human review[1]. Put the request in writing now; vetting AI vendors has the wider question list.

  3. Draft the notice and the adverse-outcome letter

    Place a short notice on the careers page and inside the application flow, and keep a template that can go out within 30 days of a rejection in which the tool played a material part[1].

  4. Name and train the human reviewer

    Pick someone with authority to change the result, give them the vendor's documentation and record each review they carry out, including what evidence they looked at.

  5. Set retention to three years

    Keep version identifiers, changelogs and the records behind each covered decision for at least three years[1]. Put the period in the firm's retention schedule so nobody purges it early.

  6. Diary the rule-making

    The Attorney General's rules on disclosures and consumer rights are due by January 1, 2027 and may add sector guidance and examples[1]. Book a policy review for the week they publish.

Example: a 12-lawyer Denver firm uses an applicant-tracking system that scores résumés for its paralegal openings. It adds a notice to the job posting, asks the vendor for its documentation, names the office manager and a partner as reviewers with power to override any score, and keeps each hiring file for three years. Its after-hours intake line, which books consultations and sends summaries to the duty lawyer, needs no change under this statute.

Enforcement, cure periods and contracts

Only the Attorney General enforces the disclosure and consumer-rights duties, and a violation is a deceptive trade practice under the Colorado Consumer Protection Act. The law creates no new private right of action. Until January 1, 2030, the Attorney General must send a notice of violation and allow 60 days to cure where a cure is possible, but not for knowing or repeated violations[1]. For 2026-27 the legislature appropriated $46,190 to the Department of Law to implement the act, based on 0.4 of a full-time position[1].

Read the liability section before you sign or renew HR software. In discrimination claims arising from a covered decision, fault is shared between developer and deployer according to their part in the violation, and a contract term that would indemnify either party for its own acts in violation of Colorado anti-discrimination law is void[1]. Using a covered tool is never a defense in itself[1]. Check indemnity and cooperation clauses in each renewal against this section, and keep the vendor's documentation with the contract.

For the wider question of where AI earns its keep in a small practice, see generative AI use cases for small law firms. For the disclosure law one state over, see Utah's AI disclosure law for law firms.

Frequently asked questions

Does Colorado's AI law apply to a law firm's client intake?

In most cases, not as intake. Legal services are not among the covered domains in SB26-189, and routine scheduling, administrative routing and customer service triage are excluded from consequential decisions[1]. The rules of professional conduct on supervision and confidentiality still apply.

We have fewer than 50 employees. Are we exempt?

Not under the new text. The relief for deployers under 50 full-time equivalent employees was in the 2024 Act[2]. SB26-189 has no headcount exemption, so coverage depends on whether you use a covered tool, for example in hiring.

Do we still have to tell Colorado callers they are speaking with AI?

SB26-189 has no stand-alone disclosure duty like the one in the 2024 Act[2]. Other states and bar opinions may require or expect disclosure, Utah among them[5], so most firms keep a short disclosure in the greeting.

When do the duties start?

The act applies to consequential decisions made on or after January 1, 2027. Developer documentation duties start the same day, and the Attorney General's rules are due by then[1].

Can a rejected job applicant sue us under the new law?

The act creates no new private right of action; the Attorney General enforces it[1]. Existing anti-discrimination claims remain available, and the act sets how fault is shared between developer and deployer in those cases[1].

References

  1. Colorado General Assembly (2026). Senate Bill 26-189, Signed Act (Automated Decision-Making Technology) (opens in a new tab) ↑

  2. Colorado General Assembly (2024). Senate Bill 24-205, Signed Act (Consumer Protections for Artificial Intelligence) (opens in a new tab) ↑

  3. Colorado General Assembly (2025). SB25B-004 (extending the effective date of SB 24-205) (opens in a new tab) ↑

  4. Colorado General Assembly (2026). SB26-189 Automated Decision-Making Technology (bill summary and history) (opens in a new tab) ↑

  5. Utah State Legislature (2025). S.B. 226 Artificial Intelligence Consumer Protection Amendments (2025, enrolled) (opens in a new tab) ↑

  6. The Florida Bar (2024). Florida Bar Ethics Opinion 24-1 (opens in a new tab) ↑

  7. U.S. Census Bureau (2023). County Business Patterns 2023: state file (cbp23st) (opens in a new tab) ↑

  8. The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) ↑

Book a demo

See Ava handle your calls.

Book a short demo. We’ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo