Colorado's AI Law Doesn't Stop at Hiring: Pay, Promotions, and Terminations Are Covered Too

In the first post in this series, I made the case that Colorado's new Automated Decision-Making Technology Act likely reaches the knockout questions in your applicant tracking system. Most of the early coverage of this law has focused on hiring, and that's where most employers stop reading.

That's a mistake. The law protects current employees, not only applicants. If software helps you decide who gets promoted, how much a raise is, who gets disciplined, or who is selected for a layoff, it may be covered starting January 1, 2027.

What Changed, and What Didn't

Colorado passed the first comprehensive state AI law in 2024 (SB 24-205). It never took effect. In 2026 the legislature repealed it and replaced it with SB 26-189, the ADMT Act.

The new law dropped the parts employers feared most: impact assessments, risk-management programs, annual reviews, and a broad duty of care to prevent "algorithmic discrimination." What it kept is a set of transparency and fairness duties:

• Notice before you use a covered tool.
• A plain-language explanation within 30 days after an adverse outcome.
• The right to access and correct personal data the tool used.
• The right to request meaningful human review.
• Three years of records showing you complied.

The narrower law is easier to comply with. It also applies to more decisions than most HR teams assume.

The Decisions It Covers

The law applies to "consequential decisions," which in employment means decisions about a person's access to, eligibility for, selection for, or compensation in a job. Employment-law commentators read that to include hiring, screening, promotion, compensation, discipline, and termination. The people it protects include job candidates and current employees.

Here is where that shows up in a small or mid-sized company:

• Compensation. A comp platform that recommends offer amounts or merit increases based on market data and performance inputs.
• Promotion. A talent tool that scores or ranks internal candidates for an open leadership role.
• Performance and discipline. A system that scores productivity, attendance, or call quality and flags employees for coaching or a performance plan.
• Termination and layoffs. A tool that ranks employees for selection in a reduction in force.

Each of these produces a score, ranking, or recommendation that feeds a decision about someone's job or pay. That is the core of the law's definition.

What Isn't Covered

Not every AI tool your team uses triggers these duties. The law excludes tools that only summarize, organize, translate, draft, route, or present information for a person to review. It also excludes spreadsheets that don't use machine learning and still require human analysis, along with calculators and databases.

That means an AI note-taker in an interview, a tool that drafts your job descriptions, or a chatbot that answers benefits questions doesn't require an employee-facing disclosure. You still need an internal AI use policy so employees know what's allowed. That point came up at an HRCI webinar I attended this week, and it's the right call.

A Human in the Loop Doesn't Settle It

Many employers assume that as long as a manager makes the final call, the software doesn't count. The Attorney General's draft rules reject that assumption.

One draft example describes a hiring manager filling a leadership role from internal candidates. A tool scores each candidate using five years of reviews, work samples, and project results. The manager sees the scores, interviews everyone, and picks the highest scorer. The draft rules presume the tool materially influenced the decision. The employer can rebut that presumption only with evidence that the decision rested on the manager's independent judgment. A version of the example where the tool ranks candidates from "strongest" to "weakest" without a number gets the same presumption.

If a manager sees the score before deciding, plan as if the tool is covered. Document how the manager reached the decision, separate from the tool's output.

What You Owe an Employee

The duties are the same ones applicants get. Give notice before the tool is used. If the outcome is adverse, provide a written explanation within 30 days that covers the decision, the tool's role, the human involvement, and the principal reasons. Generic references to company policy won't meet the standard in the draft rules. Then honor requests to correct data and to have the decision reviewed by a trained person with authority to change it.

"Adverse outcome" includes a decision that denies or ends employment, reduces compensation, or results in materially less favorable terms. One question the final rules need to answer: does a below-average merit increase count? Until they do, treat a tool-driven pay decision that leaves someone worse off than their peers as a potential adverse outcome.

Your Vendor Is Not Your Shield

The law splits responsibility between the "developer" that builds the tool and the "deployer" that uses it. If you use the tool to make decisions about your people, you are the deployer.

Three points matter for small and mid-sized employers:

• You carry your own liability. Developers and deployers can both be liable under the Colorado Anti-Discrimination Act for discrimination tied to decisions a covered tool materially influenced, with fault allocated by each party's responsibility.
• Indemnity has limits. The law voids contract clauses that indemnify a party for its own discriminatory acts. A vendor contract can't absorb your share.
• Developers owe you documentation. That includes the tool's intended uses, known limitations, data categories, and instructions for monitoring and human review. You need it to write accurate notices and explanations. Ask for it now.

Enforcement

Only the Colorado Attorney General enforces the ADMT Act, under the Colorado Consumer Protection Act. Employees and applicants can't sue under it. Until January 1, 2030, the AG must give a 60-day opportunity to cure before enforcing, unless a violation was knowing or repeated.

The cure period isn't a reason to wait. Discrimination claims under existing law don't come with one, and the records this law requires are the same records you'd need to defend those claims.

Next in This Series

The final post is the playbook: how a small or mid-sized employer inventories its tools, works with vendors, drafts notices and adverse-outcome explanations, and sets up human review before January 1.

This post is general information, not legal advice. Have employment counsel review how the law applies to your tools and processes.