AI Performance Review & Promotion Algorithm Discrimination 2026
Every major algorithmic-hiring law — NYC Local Law 144, Illinois's video interview act, most state bias-audit bills — stops at the point of hire. But the same AI scoring logic increasingly drives who gets promoted, who gets a low rating, and who gets flagged for a layoff list. That gap between compliance attention and actual use is where the next wave of discrimination claims is heading.
Compliance Attention Stopped at the Hiring Screen
NYC Local Law 144's Automated Employment Decision Tool (AEDT) definition covers tools used to "screen candidates for employment or promotion." Illinois's AI Video Interview Act covers video interviews. Most state bias-audit bills follow the same pattern: they were written in response to resume-screening and interview scoring controversies, and their text reflects that narrow origin.
Meanwhile, the same category of AI tool — a model that ranks or scores people using historical data patterns — is now embedded in performance management software that drives ongoing employment decisions: quarterly ratings, promotion readiness scores, 9-box grid placement, compensation-adjustment recommendations, and increasingly, layoff or termination risk scoring. None of the current AI-specific hiring laws clearly reach these tools, but the underlying discrimination law does.
Where the Existing Legal Exposure Actually Sits
Title VII, ADEA, and ADA Still Apply
NO AI CARVE-OUTFederal anti-discrimination law applies to any employment practice with disparate impact on a protected class — race, sex, age, or disability — regardless of whether a specific AI bias-audit statute covers the tool used to make the decision. An AI performance-scoring tool that systematically under-rates older workers or a protected group is exposed under the ADEA and Title VII today, audit law or not.
The absence of an AI-specific statute doesn't remove liability; it removes the compliance paper trail (a documented bias audit) that could otherwise help an employer defend a challenged practice.
Colorado SB 205 Reaches Beyond Hiring
BROADEST CURRENT COVERAGEColorado's AI Act defines 'consequential decisions' to include employment decisions generally, not just hiring — covering AI used for performance evaluation, promotion, compensation adjustment, and termination for Colorado-based employees. Developers and deployers face disclosure, impact-assessment, and consumer-notice obligations for these tools.
This makes Colorado the current bellwether: employers using AI performance tools on Colorado employees should treat SB 205's obligations as applicable now, not just to hiring workflows.
EEOC Guidance Extends the Same Framework Post-Hire
GUIDANCE, NOT STATUTEEEOC guidance on AI and employment discrimination frames the analysis around 'employment decisions' broadly, explicitly including performance evaluation and termination alongside hiring and promotion. The four-fifths rule and disparate-impact analysis the EEOC applies to hiring tools apply identically to a performance-scoring algorithm.
EEOC guidance isn't binding law, but it signals how the agency will investigate a discrimination charge involving an AI performance tool — and charges are already being filed on this basis.
Where AI Performance Scoring Shows Up in Practice
- Continuous performance-management platforms — AI that aggregates peer feedback, task completion data, and manager notes into a single performance score or trend line
- Promotion-readiness models — AI that flags which employees are "ready now" versus "needs development" based on historical promotion patterns, which can encode the demographic skew of who was promoted in the past
- 9-box / stack-ranking automation — AI that auto-places employees into performance-potential grids, reducing manager discretion but also reducing the ability to catch and correct a biased pattern before it's applied
- Attrition and termination risk scoring — AI that predicts which employees are "flight risks" or recommends layoff candidates based on tenure, engagement signals, or performance history
- Compensation-adjustment recommendations — AI that suggests raise or bonus amounts based on performance data that may itself reflect prior biased evaluation
Compliance Checklist for HR Tech and Employers
Monitor your HR tech product for compliance exposure
RatedWithAI helps tech teams track compliance requirements across their AI-powered products. Start with a free scan to understand your current compliance posture.
Scan Your Product for Free →Frequently Asked Questions
Does NYC Local Law 144 apply to AI performance review tools?
Only if the tool is used specifically for a promotion decision — the AEDT definition covers hiring and promotion screening, not ongoing performance ratings, termination scoring, or compensation decisions once someone is already in a role.
Can we be sued over a biased AI performance tool even with no bias-audit law covering it?
Yes. Title VII, the ADEA, and the ADA apply to any employment practice — including performance ratings and termination decisions — with a disparate impact on a protected class, regardless of whether an AI-specific bias-audit statute exists for that use case.
Does Colorado's AI Act cover promotion and termination decisions?
Yes. Colorado SB 205 defines 'consequential decisions' to include employment decisions broadly, covering AI used for performance evaluation, promotion, compensation, and termination — not just the initial hiring decision.
What's the safest way to use AI in performance management right now?
Run a voluntary disparate-impact analysis on any AI scoring tool used for promotion, compensation, or termination, keep a human in the loop with real override authority, and document your review process — even without a specific mandate, this creates the same evidentiary record a bias audit would if a claim is ever filed.
Is stack ranking or a 9-box grid itself illegal if it's AI-assisted?
No — forced ranking and 9-box methodologies are legal on their own. The risk arises when the AI scoring behind the placement systematically disadvantages a protected group, which is a disparate-impact question that depends on your specific data and outcomes, not on the methodology's name.