Using ChatGPT for Performance Reviews and Layoffs: State AI Laws, Vendor Risk & Red Lines

In June 2025, ResumeBuilder polled n = 1,342 US managers via Pollfish on how they handle promotions, raises, and headcount cuts (ResumeBuilder, Jun 2025). Over half reported using artificial intelligence for people decisions. Within that AI-using manager group, 66% of respondents used AI to decide layoffs, 64% of managers used it for terminations, 21% of respondents often or always let AI make the final decision without human input, and 24% of managers had zero formal training. On September 30, 2026, California Governor Gavin Newsom signed Senate Bill 947, the "No Robo Bosses Act," barring sole reliance on an Automated Decision System (ADS) to discipline or fire an employee and setting a $500 civil penalty per violation (Reed Smith, Oct 2026). Between an untrained supervisor pasting reviews into a chat tab and state statutes sits an expensive liability trap.

Headlines distort both AI-linked layoffs and employment law. Challenger, Gray & Christmas logged 54,836 US job cuts citing AI in 2025, 4.5% of 1.2 million total cuts in that report (Challenger PDF, Jan 2026). Pew Research (n = 5,273 US workers) found 52% of US workers worried about workplace AI (Pew, Feb 25, 2025), and Gartner survey data found 26% of candidates trust AI evaluations (Personnel Today / Gartner).

This guide covers US federal and state employment law as of October 29, 2026 across California, Illinois, Colorado, New York City, Connecticut, and Massachusetts. Because ed.cafe at No Worker Left Behind is a Boston 501(c)(3) workforce nonprofit rather than a law firm, review specific layoffs, notices, and vendor contracts with employment counsel. See also the two meanings of AI safety at work and whether Trump's AI executive order cancels state AI laws.

Why pasting performance reviews into ChatGPT triggers Automated-Decision System laws

Managers often assume employment AI laws regulate only enterprise HR software. In reality, statutory Automated-Decision System (ADS) definitions cover any computational process, including a chatbot prompt, that scores, ranks, or recommends workers for raises, discipline, or layoffs. Once a manager asks ChatGPT, Claude, or Copilot "which three people on this sheet should I cut?", that prompt becomes an unvalidated employment selection procedure.

Under Title VII, the Age Discrimination in Employment Act (ADEA), and the Americans with Disabilities Act (ADA), any selection tool producing disparate impact must be job-related and consistent with business necessity. Ad-hoc chatbot ranking fails immediately.

Picture weighing export freight pallets on an uncalibrated bathroom spring scale before signing customs forms. Even that analogy understates the risk: a bent spring errs consistently, whereas a language model shifts rankings when row order changes, penalizes protected Family and Medical Leave Act (FMLA) or disability leave gaps, repeats biased prior notes, and lacks psychometric validation. Pasting reviews into a personal Free or Plus tab also leaks personnel files (what not to paste into an AI tool at work; can your employer see your ChatGPT, Copilot, or Claude history).

Run the math on a 45-person California or Illinois firm cutting six roles (assumptions stated). Two leads paste 20 evaluations, hire dates, and attendance counts into a consumer chatbot, ask it to rank the bottom 30% of staff in that group, and delete the chat:

1. Proxy Disparate Impact: Unstructured attendance and tenure inputs steer cuts against a 56-year-old bookkeeper (ADEA) and an analyst returning from medical leave (ADA / FMLA).
2. Missing Notices: In Illinois (HB 3773, live January 1, 2026), using AI in discharge without employee notice violates civil-rights law.
3. Destroyed 4-Year Records: Both California FEHA ADS rules (live October 1, 2025) and Illinois HB 3773 require four-year retention of ADS inputs, outputs, and criteria. Deleting the chat destroys statutory records.

What California's FEHA ADS rules, SB 947, and SB 951 require of managers and HR

California enforces three overlapping employment AI mandates across hiring, reviews, discipline, and layoffs, starting at five employees under Fair Employment and Housing Act (FEHA) rules, including remote California staff.

1. California Civil Rights Council FEHA ADS Rules (In Force October 1, 2025)

Effective October 1, 2025, FEHA ADS rules bind every employer with five or more employees (Mayer Brown, Aug 2025):

• Broad ADS Scope: Covers any computational process making or facilitating hiring, promotion, evaluation, or discharge decisions.
• Direct Agent Liability: Extends FEHA liability directly to third-party screening vendors.
• Anti-Bias Testing: Weighs proactive anti-bias audits and fixes in discrimination claims and defenses.
• Tone & Affect Accommodations: Employers must accommodate applicants when an ADS measures vocal tone, facial expression, speech, or reaction time.
• Four-Year Retention: Employers shall preserve ADS inputs, outputs, criteria, and audit logs for four years.

2. California SB 947: "No Robo Bosses Act" (Signed Sept 30, 2026; Operative July 1, 2027)

Signed September 30, 2026, SB 947 adds Labor Code Part 5.5.5 (Section 1520 et seq.), operative July 1, 2027 per statutory trackers (Reed Smith, Oct 2026; LumiChats):

• No Sole Reliance: Employers shall not rely solely on an ADS to discipline, fire, or deactivate a worker.
• Human Corroboration: When primarily relying on ADS output for discipline or discharge, a human must independently review and corroborate the output against non-AI work samples, evaluations, and personnel files.
• Written Notice & Data Right: Employers primarily using an ADS for discipline or discharge must issue standalone written post-decision notice naming the human reviewer contact and the right to request underlying data.
• Penalties: Enforced by the Labor Commissioner and public prosecutors with a $500 civil penalty per violation, injunctions, and fees, plus watched PAGA exposure.

3. California SB 951 & State AI-WARN Rules (Effective January 1, 2027)

From January 1, 2027, California SB 951 mandates AI disclosure in Cal-WARN mass-layoff notices (StartBusinessByState). Connecticut SB 5 likewise requires AI disclosure in state WARN filings from October 1, 2026 (via secondary summaries; verify statutory text before filing: Gunderson Dettmer).

How Illinois, Colorado, New York City, and Massachusetts regulate AI people decisions

Binding laws in Illinois, Colorado, New York City, and Massachusetts already penalize secret AI screening, deceptive vendor claims, and automated honesty scoring. Federal Executive Order 14365 (December 11, 2025) has invalidated none of these state statutes (Sidley, Dec 23, 2025).

Watch four multi-state traps:

1. Illinois HB 3773 (Live Jan 1, 2026): Notify staff and applicants whenever AI touches recruitment, hiring, promotion, renewal, training, discharge, discipline, or tenure, avoid discriminatory ZIP-code proxies, and keep records four years (Hinshaw, 2026).
2. Colorado SB 26-189 (Effective Jan 1, 2027): Signed May 14, 2026 (replacing the old June 2026 date), SB 26-189 mandates pre-notices, a 30-day adverse explanation, human review, three-year records, and voids vendor indemnity clauses shifting bias liability onto employers (Legal 500).
3. NYC Local Law 144: A December 2025 State Comptroller audit flagged 17 likely violations across reviewed complaints where the city agency had found only 1 (NY OSC, Dec 2025).
4. Massachusetts Lie-Detector Statute (M.G.L. c. 149, s. 19B) & Chapter 93A: M.G.L. c. 149, s. 19B bars subjecting any worker or applicant to "any device, mechanism, instrument or written examination" used to assess honesty, mandates warning wording on every Massachusetts job application, and sets a $500 minimum fine plus treble lost-wage damages and fees (M.G.L. c. 149 s. 19B). In Baker v. CVS Health Corp. (D. Mass.), a claim over HireVue's AI video-interview integrity scoring survived dismissal before settling, though later Massachusetts rulings narrowed s. 19B where tools measure job skills rather than truthfulness (Chambers / Sidley, 2026). The Massachusetts AG also enforces c. 151B and c. 93A against untested AI models, shown by her $2.5 million Earnest settlement (Mass.gov, Jul 10, 2025; Orrick, Apr 2024).

Why Mobley v. Workday ended the "blame the software vendor" defense

Employers cannot assume a brand-name HR platform insulates them from bias suits. In Mobley v. Workday, Inc. (Case No. 3:23-cv-00770, N.D. Cal.), Judge Rita F. Lin held that screening vendors acting as gatekeepers face direct "agent" liability while employers remain liable as principals for automated filters enabled in their tenant.

Derek Mobley, a Black applicant over 40 with anxiety and depression, alleged rejection from 100+ jobs across employers using Workday screening tools. Three rulings reshaped HR procurement:

1. Statutory "Agent" Theory (July 2024): Because Title VII, the ADA, and the ADEA define "employer" to include "any agent," a vendor delegated authority to score and reject applicants acts plausibly as a statutory agent (Maynard Nexsen).
2. Nationwide ADEA Collective (May 2025): The court preliminarily certified a nationwide ADEA collective action for applicants 40+ screened by Workday recommendation tools (Holland & Knight, May 2025).
3. HiredScore Scope & June 22, 2026 Order: Later orders kept HiredScore tools in scope, and on June 22, 2026, the court granted in part and denied in part a motion to dismiss while preserving core agency liability (Clearinghouse Case 44074).

Boilerplate SaaS contracts cap vendor liability at 12 months of fees and force the employer to indemnify the vendor. Without contract redlines (and rules like Colorado SB 26-189), buyers fund defense costs for opaque vendor weights. Test tools up front with our 50-example spreadsheet method and set internal rules in our one-page AI use policy.

Safe vs prohibited uses: what managers may draft with AI and what they must decide

Separate polishing human-decided prose from delegating evaluation, ranking, or credibility scoring:

• Safe (Green Zone, approved zero-training work account): Polishing grammar on a review after the manager picks the rating and types verified project facts; drafting job-sample interview rubrics up front; cleaning up role descriptions.
• Conditional (Yellow Zone, HR review + notices + 4-year logs): Audited ATS checks on objective licenses, with Illinois, NYC, Colorado, or California notices, accommodations, and 4-year archives active.
• Prohibited (Red Zone): Asking ChatGPT, Copilot, or Claude to rank staff for layoffs, promotions, PIPs, or raises; pasting raw peer comments, medical/FMLA notes, or pay sheets into consumer bots; citing unverified AI meeting notes in discipline (AI notetakers in meetings; how often AI is wrong); or enabling video tone, affect, or honesty scoring (M.G.L. c. 149, s. 19B).

The ed.cafe Red-Line Card, Corroboration Log, and 12 Vendor Questions

The ed.cafe People-Decision AI Red-Line Card, Corroboration Log & 12 Vendor Questions, published by No Worker Left Behind, gives managers and HR buyers a desk red-line card, a California SB 947 / 4-year corroboration log, and 12 post-Mobley vendor contract asks.

THE ED.CAFE PEOPLE-DECISION AI DEFENSE PACK (LATE-2026 EDITION)
Review with employment counsel before reductions in force or HR SaaS renewals.

PART 1: MANAGER PEOPLE-DECISION RED-LINE CARD
Workflow            | SAFE (Approved Work Account Only)   | RED LINE (Never Do This)
--------------------+-------------------------------------+-------------------------------------
Performance Reviews | Polish YOUR dated, verified notes.  | Ask AI what rating staff deserve.
Layoffs / Firing    | Edit transition FAQs AFTER human    | Ask AI to rank staff or pick cuts
                    | selection is signed off.            | (barred under CA SB 947 sole rule).
Discipline / PIPs   | Check clarity on human-written PIPs.| Cite unverified AI meeting notes.
Hiring & Interviews | Draft job-sample rubrics up front.  | Score video tone/honesty (MA s.19B).

MANDATORY MASSACHUSETTS APPLICATION NOTICE (M.G.L. c. 149, s. 19B(2)(b)):
Include verbatim on every Massachusetts employment application:
"It is unlawful in Massachusetts to require or administer a lie detector test as a
condition of employment or continued employment. An employer who violates this law shall
be subject to criminal penalties and civil liability."

PART 2: HUMAN CORROBORATION & 4-YEAR RECORD LOG (CA FEHA / SB 947 / IL HB 3773 / CO SB 26-189)
Complete BEFORE any adverse hiring, promotion, disciplinary, or termination action where
any software score, filter, or AI tool touched the workflow. Retain for 4 YEARS.

1. Action: [ ] Hiring  [ ] Promotion  [ ] Discipline/PIP  [ ] Termination/Layoff
2. Worker State(s): ____________________  Tool & Version (if any): _____________________
3. Notices & Accommodations Verified:
   [ ] IL HB 3773 AI notice delivered    [ ] NYC LL144 audit + 10-day notice verified
   [ ] CO SB 26-189 pre-notice & 30-day explanation ready (from Jan 1, 2027)
   [ ] CA SB 947 written notice & data-request contact ready (from Jul 1, 2027)
   [ ] CA SB 951 / CT SB 5 AI-WARN check done (mass layoffs)
   [ ] Disability / deaf / speech accommodation offered before timed or audio screens
4. INDEPENDENT HUMAN CORROBORATION (Non-AI Primary Records Inspected Directly):
   - Primary Record #1 (dated work sample / KPI / attendance log): _____________________
   - Primary Record #2 (signed prior review / corroborated note): ______________________
   - Checked for protected leave (FMLA/ADA/PWFA) or source errors? [ ] YES
   - Independent human decision rationale: _____________________________________________
5. 4-Year Archive Path: ______________________  Reviewer Signature & Date: _____________

PART 3: 12 HIRING-VENDOR QUESTIONS & CONTRACT ASKS AFTER MOBLEY V. WORKDAY
 1. Kill Switch: Can we disable auto-ranking, auto-reject, and match scores per tenant?
 2. Honesty & Affect Ban: Warrant zero facial, vocal-tone, demeanor, or honesty scoring
    (compliance with Massachusetts M.G.L. c. 149, s. 19B).
 3. Proxy Exclusions: Confirm ZIP codes, graduation years, school default rates, and
    protected leave gaps are excluded from model inputs.
 4. Bias Audits: Share latest independent disparate-impact audit with sample sizes (n)
    and four-fifths ratios across race, sex, age (40+), and disability.
 5. Model Change Notice: Give 30 days' written notice before weight or model updates.
 6. Free 4-Year Exports: Export inputs, criteria, and outputs in CSV/JSON at zero extra
    cost to meet CA FEHA and IL HB 3773 4-year retention rules.
 7. Individual Reasons: Output specific per-person factors for CO SB 26-189 30-day
    explanations and CA SB 947 employee data requests.
 8. Accommodation Bypass: Provide a candidate click-path for human review or captioning
    without down-ranking the applicant.
 9. Zero Model Training: Warrant in the DPA that resumes and reviews never train models.
10. Statutory Notice Copy: Supply exact product/purpose fields for IL, NYC, and CO notices.
11. Audit Cooperation: Produce validation records within 15 business days for regulators.
12. Indemnification Redline: Strike one-way employer indemnities (void in CO Jan 1, 2027);
    require vendor indemnity for disparate-impact claims caused by vendor algorithms.

What workers and applicants can ask for when AI touches a review, rejection, or layoff

Workers have real remedies when automated screens misfire. On March 19, 2025, the ACLU filed a charge alleging a deaf Indigenous Intuit employee forced to take a HireVue AI video interview for promotion was denied human captioning and penalized for her "communication style" (HR Dive, Mar 19, 2025).

Take three steps if AI touches your review, application, or layoff:

1. Request accommodations in writing up front: Cite the ADA and state rules (such as California FEHA ADS rules) to request human review, live captioning, or an untimed skills alternative.
2. Check notices and request data: Look for Illinois HB 3773 or NYC LL144 notices today, plus Colorado SB 26-189 (30-day explanation) and California SB 947 (ADS data request) rights as 2027 dates take effect.
3. Save non-confidential records: Keep signed reviews and quota sheets (zero client PII) before signing a severance release; group layoffs of staff 40+ also require OWBPA age-and-title tables. Compare workplace protocols in the ed.cafe Staff Room.

Where this stops being true

Statutory summaries hit boundaries requiring employment counsel:

• Pending agency rules: Final Illinois IDHR rules, Colorado AG rules due January 1, 2027, and California SB 947 PAGA rulings can refine notice fields.
• Narrowed Massachusetts s. 19B scope: After Baker v. CVS, Massachusetts courts declined to extend M.G.L. c. 149, s. 19B to tools measuring job skills rather than honesty (Chambers / Sidley, 2026).
• Union contracts: CBAs often bar unilateral AI evaluation or layoff tools (Allwork, Aug 2026).
• Federal contractors & civil service: OFCCP validation rules and merit exams carry separate mandates.

Questions people ask

Key terms

• Automated-Decision System (ADS): Under California FEHA rules (live Oct 1, 2025), any computational process making or facilitating an employment decision, including chatbot ranking.
• Human Corroboration (CA SB 947): California rule (operative Jul 1, 2027) requiring human verification of ADS-backed discipline or discharge against non-AI records.
• Statutory Agent Theory (Mobley v. Workday): Doctrine holding AI screening vendors directly liable alongside employers under Title VII, ADA, and ADEA.
• Illinois HB 3773: Illinois law live Jan 1, 2026 mandating AI employment notices and four-year record retention.
• Colorado SB 26-189: May 2026 overhaul live Jan 1, 2027 requiring pre-notice, 30-day explanations, and voiding vendor liability-shifting indemnities.
• Massachusetts Lie-Detector Statute (M.G.L. c. 149, s. 19B): Massachusetts ban on devices assessing honesty, backed by $500 fines and treble wages.
• Disparate Impact: Unlawful practice where a neutral tool disproportionately screens out a protected group without job-related business necessity.

Sources

1. ResumeBuilder, "Managers Using AI for Promotions & Firing (n = 1,342)" (Jun 2025): ResumeBuilder
2. Mayer Brown, "CA Employment AI Rules Effective Oct 1, 2025" (Aug 2025): Mayer Brown
3. Reed Smith, "CA SB 947 No Robo Bosses Act" (Oct 2026): Reed Smith
4. StartBusinessByState, "CA SB 951, CO SB 26-189 & IL HB 3773" (2026): StartBusinessByState
5. Hinshaw, "Illinois AI Regulations (HB 3773)" (2026): Hinshaw
6. Reinhart / Legal 500, "Colorado SB 26-189" (May 2026): Legal 500
7. NY State Comptroller, "NYC LL144 Audit" (Dec 2025): NY OSC
8. Mass. General Court, "M.G.L. c. 149, s. 19B": malegislature.gov
9. Chambers / Sidley, "AI 2026: Massachusetts (Baker v. CVS)": Chambers
10. Orrick, "MA AG AI Guidance (c. 93A & c. 151B)" (Apr 2024): Orrick
11. Mass.gov AG, "$2.5M Earnest AI Settlement" (Jul 10, 2025): Mass.gov
12. Clearinghouse, "Mobley v. Workday (Jun 22, 2026 Order)": Clearinghouse
13. Maynard Nexsen, "Mobley v. Workday Agent Liability": Maynard Nexsen
14. Holland & Knight, "Mobley ADEA Collective" (May 2025): Holland & Knight
15. Challenger, "2025 Job Cuts (54,836 AI Cuts)" (Jan 2026): Challenger PDF
16. Pew Research, "US Workers on AI (n = 5,273)" (Feb 25, 2025): Pew
17. Personnel Today / Gartner, "Candidate AI Survey (26%)" (2025): Personnel Today
18. HR Dive, "ACLU Charge: Intuit & HireVue" (Mar 19, 2025): HR Dive
19. Gunderson Dettmer, "CT SB 5 AI WARN" (Spring 2026): Gunderson
20. Sidley, "Executive Order 14365" (Dec 23, 2025): Sidley
21. Allwork, "Union Bargaining & AI Rules" (Aug 2026): Allwork
22. LumiChats, "CA SB 947 Explained" (2026): LumiChats