Skip to main content
AI & Automation

California and Colorado AI rules from January 2027: a checklist for businesses using AI agents

By Prilient ·

California and Colorado AI rules from January 2027: a checklist for businesses using AI agents
Short answer: from 1 January 2027, California and Colorado regulate software, including AI agents, that decides or strongly shapes decisions about jobs, loans, housing, insurance, education or health care. California requires a pre-use notice, an opt-out or human appeal, access requests and risk assessments. Colorado requires a pre-use notice, an explanation within 30 days of an adverse outcome, and human review on request. Buying the AI from a vendor does not move these duties to the vendor.

This is general information, not legal advice.

Two sets of rules start on the same day. The California Privacy Protection Agency (CPPA) finalised regulations on automated decisionmaking technology (ADMT), and compliance with the ADMT parts begins on 1 January 2027 (Bass, Berry & Sims). Colorado's governor signed SB 26-189 on 14 May 2026, replacing the state's 2024 AI Act with a narrower law that also takes effect on 1 January 2027 (Colorado Governor's Office). If an AI agent screens your job applicants, borrowers or tenants, this guide covers what counts, what to set up and what to ask your vendor.

The two rules at a glance

California (CCPA ADMT rules) Colorado (SB 26-189)
Starts 1 January 2027 1 January 2027
Who must comply Businesses covered by the CCPA Anyone doing business in Colorado that uses covered ADMT
Decisions covered Lending, housing, education, employment, compensation, health care Education, employment, housing, lending, insurance, health care, essential government services
Before use Pre-use notice Clear and conspicuous notice
People's rights Opt out (or human appeal), access to information Explanation within 30 days of an adverse outcome, data correction, human review
Paperwork Risk assessments; information to the CPPA by 1 April 2028 Records kept for three years
Enforcement CPPA and California Attorney General Colorado Attorney General; 60-day cure period

What California requires

The rules apply to businesses covered by the California Consumer Privacy Act. That generally means for-profit businesses doing business in California with annual gross revenue above $25 million, or that buy, sell or share the personal information of 100,000 or more California consumers or households, or that earn half their revenue from selling or sharing personal information (National Law Review).

ADMT means technology that processes personal information and uses computation "to replace human decisionmaking or substantially replace human decisionmaking". A "significant decision" is one about financial or lending services, housing, education, employment or independent contracting, compensation, or health care services. Targeted advertising on its own is excluded.

From 1 January 2027, a business using ADMT for a significant decision must:

  • give a clear pre-use notice at or before the point of collection, explaining the purpose, how the ADMT uses personal information, and the person's rights;
  • let people opt out through at least two methods, unless an exception applies, such as offering a human appeal instead;
  • answer access requests about the ADMT's purpose, its output and whether a human reviewed it, acknowledging within 10 business days and responding within 45 days;
  • carry out a written risk assessment, review it at least every three years and update it within 45 days of a material change.

Information on risk assessments done in 2026 and 2027 goes to the CPPA by 1 April 2028, with an attestation signed by a named person.

What Colorado requires

SB 26-189 applies to anyone "doing business in Colorado" that uses a covered ADMT, which the law calls a deployer (Seyfarth). The tool is covered when its output "materially influences" a consequential decision: a non-trivial factor that changes the outcome, including by "constraining, ranking, scoring, recommending, classifying" (Lathrop GPM). Consequential decisions cover education, employment, housing, financial or lending services, insurance, health care and essential government services.

A deployer must:

  • give clear and conspicuous notice before using the tool, with a way to learn more; a prominent public notice where people interact with you can satisfy this;
  • within 30 days of an adverse outcome, explain the decision and the tool's role in plain language, and say how to ask for more information;
  • on request, explain how to correct inaccurate personal data and offer meaningful human review and reconsideration, as far as commercially reasonable;
  • make notices accessible to people with disabilities and limited English, and keep records for three years.

The reviewer must be trained, able to override and must not "default to system output". The Colorado Attorney General enforces the law as a deceptive trade practice. There is no private right of action, and businesses generally get 60 days to cure a violation unless it was knowing or repeated.

Do AI chatbots and AI agents count?

It depends on the job, not the label. An agent that only answers questions is generally outside both rules: Colorado excludes conversational tools not intended for consequential decisions, scheduling, customer service triage and marketing.

An agent that scores, ranks or filters people is different. In Colorado, ranking applicants is named in the definition of "materially influence", so an interview-screening agent that shortlists candidates is likely covered (Ogletree). In California, the test is whether a person is truly involved. The reviewer must know how to read the output, consider other relevant information, and have the authority to make or change the decision. A recruiter who approves every AI recommendation unread probably does not.

Which AI agent jobs are likely in scope

Based on the definitions above; your answer depends on how each agent is set up.

Agent job Likely in scope? Why
Interview screening that scores or shortlists candidates Likely yes in both Employment is covered, and Colorado names ranking and scoring
Pre-screening loan, tenancy or insurance applications Likely yes in both Lending, housing and insurance are covered areas
Debt collection that decides who is offered a payment plan or escalated Check carefully Can shape access to financial services; some credit decisions fall under federal lending rules instead
Customer support that answers questions Usually not Colorado excludes conversational tools not meant for consequential decisions
Receptionist or appointment setter that books slots Usually not Colorado excludes scheduling; check if it decides who gets health care first
WhatsApp or web lead qualification Usually not Marketing is excluded; in scope if it screens people out of a loan, home or policy

Your vendor does not carry the duty for you

Both laws put the main duties on the business that uses the tool. Colorado also requires developers to give deployers the tool's intended uses, training data categories, known limitations, human-review instructions and what deployers need for their notices (Lathrop GPM).

Ask every AI vendor these six questions before 1 January:

  1. Which decisions can your tool make on its own, and which does it score, rank or recommend?
  2. Which kinds of personal information does it use for each decision?
  3. Can our staff see why it reached an output, and override it, with each override logged?
  4. Will you give us written documentation we can use for notices and risk assessments?
  5. Can the tool record an opt-out or a request for human review and route it to a person?
  6. Will you tell us before an update changes what the tool decides or the data it uses?

An eight-step checklist before 1 January 2027

  1. Check whether each law covers you. Compare yourself with the California thresholds and ask whether you do business in Colorado; being based outside the US does not settle it.
  2. List every AI tool that touches decisions about people, including vendor features inside your HR, CRM or lending software.
  3. Mark which ones decide or rank in a covered area: hiring, pay, credit, tenancy, insurance, education or health care.
  4. Decide where a person truly decides. If you rely on human review, make sure reviewers are trained, see the evidence and can override the tool.
  5. Write the pre-use notice. Put it where people meet the tool: the job application, the loan form, the first chat message.
  6. Set up the request routes. Two opt-out methods for California, an appeal or human-review route for both states, and an access-request process that meets the 10-day and 45-day deadlines.
  7. Prepare the adverse-outcome explanation: a plain-language template per decision type, sent within Colorado's 30 days.
  8. Document it. Run the California risk assessment, keep Colorado records for three years, and review both whenever an agent's job changes.

How Prilient can help

Prilient builds custom AI agents and runs six ready-made Prilient AI Agents: Receptionist, Customer Support, WhatsApp Lead Generation, Appointment Setter, Debt Collection and HR Interview Screening. Since 2019 the team has delivered 500+ projects for 50+ clients. Through AI consulting, we help you map which agents decide, which only assist, and where a person must stay in the loop.

Book a free 30-minute call. We will go through the agents you run or plan against both rules, and list the questions to put to your vendor and your lawyer.

Questions about the California and Colorado AI rules

When do the California and Colorado AI decision rules start?

Both start on 1 January 2027. California's CCPA rules on automated decisionmaking technology and Colorado's SB 26-189 apply to software that makes or strongly shapes decisions about jobs, lending, housing, education, health care and, in Colorado, insurance.

Do these rules apply to a company based outside the US?

They can. California's rules follow the CCPA, which covers businesses doing business in California that meet its revenue or data thresholds. Colorado's law covers anyone doing business in Colorado that uses a covered tool. Ask your lawyer how they apply to you.

Does a recruiter reviewing the AI's shortlist take us out of scope?

Not automatically. In California, the reviewer must know how to read the output, consider other information and have authority to change the decision. In Colorado, a tool that ranks or scores candidates can still materially influence the decision.

Do AI chatbots have to say they are AI under these rules?

Not as a general rule. Neither law adds a standalone duty to say a chatbot is an AI, and Colorado excludes conversational tools not meant for consequential decisions. Other laws may still require disclosure, so check with your lawyer.

Our AI tool comes from a vendor. Who is responsible?

Your business is. Both laws put the main duties on the business that uses the tool. Colorado also requires developers to give you documentation on intended uses, limitations and human review, so ask for it in writing.

What happens if we do not comply in Colorado?

The Colorado Attorney General enforces SB 26-189 as a deceptive trade practice. There is no private right of action, and businesses generally get 60 days to fix a violation after notice unless it was knowing or repeated.

Leave A Reply

Your email address will not be published. Required fields are marked *

Have a Project in Mind?

Let us help you build something amazing. Get in touch with our team today.