Almost every conversation about AI in compensation is about what it can do for you. Faster benchmarking. Continuous pay equity monitoring. Predictive modeling. Merit recommendations that arrive before the manager finishes their coffee.
That conversation is worth having.
There is a second conversation getting far less airtime, and if you are a compensation leader or a CHRO, it is the one that can hurt you. What happens when the AI gets a compensation decision wrong? Who has to explain it? And who pays?
I have been in HR technology for four decades, and I have watched a lot of software categories grow up. Early adopters may feel some advantage, but they also absorb the risk. In most categories that risk is operational: bugs, missing features, integration headaches. With AI in compensation decisions, the risk is legal. And the law has stopped treating compensation as somebody else’s problem.
Your Compensation Decisions Are Now in Scope
For the last few years, the AI employment rules most people heard about were hiring rules. That is no longer true.
Colorado is the clearest example. In May 2026, Colorado replaced its original AI law with a new one, Senate Bill 26-189. It takes effect January 1, 2027. The new law covers automated decision-making technology that materially influences “consequential decisions” in employment, and compensation is written into that definition. If an AI tool is more than a minor factor in deciding someone’s compensation, the law reaches it.
Here is what Colorado will require of an employer using one of those tools:
- Notice before you use it. A clear and conspicuous notice to every employee whose compensation the tool will influence, before it is used.
- An explanation after an adverse outcome. Within 30 days, a plain-language description of the decision and the role the technology played in it.
- Correction and human review. On request after an adverse outcome, instructions for correcting inaccurate personal data and, where commercially reasonable, an opportunity for meaningful human review and reconsideration.
- Records. At least three years of compliance documentation.
The attorney general enforces it.
Colorado is not alone, and it is not even first.
California has had regulations in effect since October 1, 2025 that bar discriminatory use of automated decision systems in hiring, promotion, training, pay, benefits, leave, and termination. Pay is named outright. Employers have to keep automated decision system data for four years, and the regulations treat an employer’s agents as employers too, so handing the work to a vendor does not hand off the responsibility.
Illinois, since January 1, 2026, has covered employer use of AI in decisions about hiring, promotion, discipline, and “the terms, privileges, or conditions of employment.” That language reaches compensation. The state withdrew its draft notice rules in June, but the law itself is in force.
What That Looks Like in a Real Merit Cycle
Picture a merit cycle in early 2027. You have rolled out an AI tool that recommends increases, and your managers mostly accept what it suggests. A few hundred of your employees work in Colorado.
Before the cycle opens, those employees are owed notice that an automated tool will help decide their compensation. Later, when the comp statements go out, some of them will get less than they expected. Is a below-target increase an “adverse outcome” under the law? That is a question you want your counsel to answer before the cycle opens, not after. If the answer is yes, each of those employees is owed, within 30 days, a plain-language description of the decision and the role the tool played in it. They can ask you to correct the data it used, and they can ask a person to reconsider.
Now ask yourself honestly. Could you write that explanation? For every one of them? Could your vendor even tell you what role the model played, in words an employee would accept?
Here is the part that should keep a comp leader up at night. A hiring tool touches applicants once. A compensation tool touches every employee, every cycle. Those employees know your name, they know where HR sits, and they compare notes.
The Courts Are Already Building the Case Law
The litigation so far is about hiring, because that is where AI got adopted first. Read these cases anyway. The legal theories being tested on hiring tools today are the ones your compensation tools will face next.
EEOC v. iTutorGroup (settled 2023). iTutorGroup agreed to pay $365,000 to settle an EEOC lawsuit alleging its application software automatically rejected women 55 and older and men 60 and older. The EEOC said more than 200 qualified U.S. applicants were turned away. The age cutoffs were programmed in. It is widely cited as the EEOC’s first settlement involving an automated hiring tool, and it tells you the agency is willing to bring these cases.
Kistler v. Eightfold AI (filed January 2026). Job seekers filed a class action alleging Eightfold scraped personal data on more than one billion workers and used it to score and rank applicants. It is not a bias case. It is brought under the Fair Credit Reporting Act, on the theory that a secret score built from compiled personal data is a consumer report, with all the disclosure and dispute rights that come with one. The case is early and the allegations are unproven. But it shows plaintiffs’ attorneys are building new theories, not just recycling discrimination claims.
Mobley v. Workday (filed 2023). This is the one to read closely, and it keeps getting bigger. The plaintiffs allege that AI-enabled applicant screening screened people out based on age, race, and disability. In May 2025, the federal court in the Northern District of California preliminarily certified a nationwide collective of applicants aged 40 and over under the Age Discrimination in Employment Act. In June 2026, the court refused to dismiss most of the remaining claims, including disability and California state-law discrimination claims. On September 14, 2026, the plaintiffs asked the court to certify four classes: Black applicants, women, applicants over 40, and applicants with disabilities. Their filing cites more than 356 million applications submitted through the vendor’s recruiting platform in 2024 alone. The certification hearing is set for March 2027.
The vendor denies that its tools discriminate. Nobody has shown that anyone programmed discrimination in on purpose. The allegation is that the AI produced discriminatory outcomes anyway.
That is the scenario most organizations are not ready for.
The Employer Is on the Hook
When an AI tool used in HR produces discriminatory outcomes, the employer can be liable under existing employment law. Under a disparate-impact theory, that liability does not depend on anyone intending the bias. It does not matter that the vendor acted in good faith, and it does not matter that your team configured the system exactly as instructed.
Mobley added a wrinkle: the court let the plaintiffs argue that the vendor itself can be liable as the employer’s agent. California’s regulations say much the same thing. The liability may not stop with you. It still starts with you.
That makes your procurement decision a compliance decision. When you adopt an AI tool, you are also deciding which risks you are willing to absorb. Most procurement conversations never get to that level. They stay at the feature level, and the demo is very good.
The Rest of the Map (September 2026)
The rules change depending on where your employees sit, and they are still being written.
- New York City, Local Law 144. Requires an independent annual bias audit before using an automated employment decision tool for hiring or promotion. It does not reach compensation decisions, and enforcement has been thin: a December 2025 audit by the New York State Comptroller called the city’s complaint process “ineffective” and found at least 17 potential violations in a sample of companies where the city’s own review had found one. Penalties run up to $500 for a first violation and $500 to $1,500 for each one after that, and every day a noncompliant tool is in use counts as a separate violation.
- European Union, AI Act. Treats AI used to make decisions affecting “terms of work-related relationships,” promotion, termination, and performance evaluation as high risk. Compensation is inside that line. The European Commission’s own AI Act guidance says “remuneration decisions fall within the terms of a work-related relationship,” which puts AI that shapes compensation in the high-risk category. Employers using these systems must assign human oversight to people with the competence and authority to exercise it, and must inform workers before the system is used on them. In July 2026, the EU pushed the start date for these obligations to December 2, 2027. That is extra runway, not an exemption.
- European Union, Pay Transparency Directive. Member states were due to write it into national law by June 7, 2026. Only four made the deadline, and by late September only five had finished. Don’t read that as a reprieve. The EU has ruled out any delay at the EU level. Once in force, employers with 150 or more workers report gender pay gaps, and a gap of 5% or more in any category of workers that the employer can’t justify on objective, gender-neutral criteria, and doesn’t correct within six months, triggers a joint pay assessment. In a pay discrimination claim, once the employee establishes facts suggesting discrimination, the burden moves to the employer to prove there wasn’t any. If an AI tool shaped those compensation decisions, you will need to explain what it did.
Five Questions to Ask Before AI Touches a Compensation Decision
None of this means AI has no place in compensation. It means the decision deserves the same due diligence as any other decision with legal and financial exposure. Before you adopt a tool that influences compensation, get answers to these five questions, before anyone shows you a feature:
- How was the model trained, and on what data? If the training data reflects historical pay inequities, the model will carry them forward. Vendor reassurance is not documentation.
- Has an independent bias audit been done? Not a self-assessment. An independent review of outcomes across demographic groups, with documented results.
- Can you meet Colorado’s notice and explanation requirements, and California’s four-year record retention, with this tool as it ships? Ask the vendor to show you, not tell you. Then ask the same question for every other place your employees work.
- What does the indemnification actually cover? Read the liability cap and set it next to what defending an employment class action would cost.
- When an employee asks why they got the increase they got, can you explain it in terms they can evaluate? Not in model inputs or recommendation weights. In plain language.
That last one matters more than the other four. Compensation touches individual people and individual paychecks. When someone doesn’t understand why they are paid what they are paid, or believes the outcome was unfair, that conversation doesn’t happen with the software vendor. It happens with HR. In Colorado, starting in 2027, it may have to happen in writing.
The Larger Point
Compensation decisions carry legal, financial, and human weight that most software decisions don’t. The person who gets the raise, or doesn’t, is a specific individual with legal standing and a direct stake in the outcome.
The vendors leading the AI-in-comp charge are moving fast. The law moves slower, but it is moving, and it has reached compensation.
Early adopters may get a head start. They also get to be the test cases.
That is not a reason to stay out of the category forever. It is a reason to slow down, ask the hard questions first, and keep a person, not a model, accountable for every number your comp cycle produces.
Sources:
- Colorado SB 26-189, signed May 14, 2026, effective January 1, 2027. Ogletree, “Colorado’s New AI Act Targets Automated Decision-Making for Consequential Decisions” (ogletree.com); Epstein Becker Green, “Inside Colorado’s Senate Bill 26-189: Impacts and Implications for Employers” (ebglaw.com); Venable, “Colorado’s New AI Law: What Employers Need to Know About Automated Decision-Making,” July 2026 (venable.com); HR Dive, “Revamped Colorado AI law targets ‘consequential’ HR decisions, takes effect in 2027,” May 15, 2026 (hrdive.com)
- California Civil Rights Council automated decision system regulations, effective October 1, 2025. Duane Morris, “AI Catches Up: California Employers’ Regulations on Automated Decision Systems Now in Effect” (duanemorris.com)
- Illinois Public Act 103-0804 (HB 3773), effective January 1, 2026. Thompson Hine, “Illinois Enacts State Laws Regulating AI Use in Employment” (thompsonhine.com); Burke, Warren, MacKay & Serritella, “Illinois Department of Human Rights Withdraws Proposed AI in Employment Rules,” June 16, 2026 (burkelaw.com)
- EEOC press release, “iTutorGroup to Pay $365,000 to Settle EEOC Discriminatory Hiring Suit,” September 11, 2023 (eeoc.gov)
- Kistler v. Eightfold AI Inc., filed January 20, 2026. Analysis: Jones Walker, “AI Hiring Under Fire: What the Eightfold Lawsuit Means for Every Employer” (joneswalker.com)
- Mobley v. Workday, Inc., No. 3:23-cv-00770 (N.D. Cal.). Preliminary ADEA collective certification, May 16, 2025: CDF Labor Law, “Federal Court Grants Preliminary Certification in Landmark AI Hiring Bias Case” (cdflaborlaw.com). Agent-liability theory: Akin Gump AI Law and Regulation Tracker (akingump.com). Ruling on motion to dismiss Third Amended Complaint, June 22, 2026: HR Dive, “Workday can’t shake California AI discrimination claims” (hrdive.com); RPJ Law (rpjlaw.com). Motion for class certification filed September 14, 2026, hearing March 9, 2027: Reuters, as reported by Lawyer Monthly, “Workday AI Hiring Bias Case Seeks Class Status,” September 21, 2026 (lawyer-monthly.com)
- NYC Local Law 144 of 2021, NYC Admin. Code §20-870 et seq. New York State Comptroller, “Enforcement of Local Law 144: Automated Employment Decision Tools,” Report 2024-N-6, December 2, 2025 (osc.ny.gov)
- EU AI Act, Regulation (EU) 2024/1689, Annex III point 4 and Article 26. Compensation within 4(b): European Commission, AI Act Service Desk, “Employment” (ai-act-service-desk.ec.europa.eu). AI Omnibus timing: European Commission, “AI Omnibus enters into force,” July 2026 (digital-strategy.ec.europa.eu)
- EU Pay Transparency Directive, Directive (EU) 2023/970, Articles 9, 10 and 18. Transposition status: Lewis Silkin, “Pay Transparency Directive FAQs,” June 25, 2026, and “EU Pay Transparency Directive 2026: no EU-level delay,” July 1, 2026 (lewissilkin.com); Lockton, “EU Pay Transparency Directive: Implementation Status,” September 25, 2026 (global.lockton.com)
