State AI Employment Law

Illinois, Colorado, and California Diverge in 2026

Executive Summary

Illinois’ HB 3773, Colorado’s SB 26-189, and California’s FEHA regulations each govern AI hiring tools in 2026, yet none makes sole AI control a precondition for compliance.

A startup that screens resumes, scores interviews, or monitors performance with AI faces three separate regimes, each attaching its notice, testing, or review duty the moment an automated output can shape a covered decision, whether or not the software replaces a human entirely. Illinois’ amendment to the Human Rights Act, codified at 775 ILCS 5/2-102(L), took effect January 1, 2026, requiring notice whenever artificial intelligence can influence recruitment, hiring, promotion, discipline, or discharge, and separately banning zip codes as a proxy for protected class membership. California’s Civil Rights Council regulations, at 2 California Code of Regulations sections 11008 and 11009, took effect October 1, 2025, extending FEHA’s antidiscrimination duties to any automated-decision system that makes or helps make an employment decision, with a four year recordkeeping rule and expanded liability for the vendors that build or run the tool.

Colorado presents the most volatile timeline of the three. The original Colorado AI Act, Colorado Revised Statutes section 6-1-1701 and following, signed May 17, 2024, was postponed from February 1, 2026 to June 30, 2026 by SB 25B-004, then repealed before that postponed date arrived. Governor Jared Polis signed its replacement, SB 26-189, the Automated Decision-Making Technology Act, on May 14, 2026, dropping the mandatory impact assessment in favor of a narrower notice and human review model effective January 1, 2027.

The practical result for a multistate employer is a staggered compliance calendar running from October 2025 through January 2027, three different definitions of the triggering conduct, and no state that lets a vendor’s marketing claim of human oversight excuse the underlying notice duty.

Detailed Findings

What does Illinois’ HB 3773 require of employers using AI in hiring

Illinois amended the Illinois Human Rights Act through House Bill 3773, effective January 1, 2026, creating two independent violations at 775 ILCS 5/2-102(L). The first, subsection (L)(1), makes it a civil rights violation to use artificial intelligence in a way that has a discriminatory effect on the basis of a protected class, including the use of zip code as a proxy for race or national origin. The second, subsection (L)(2), separately requires an employer to notify employees and applicants when artificial intelligence is used for recruitment, hiring, promotion, renewal of employment, selection for training or apprenticeship, discharge, discipline, tenure, or the terms, privileges, or conditions of employment.

Neither provision conditions liability on the AI tool acting as the sole or primary basis for the decision. The statutory language reaches any artificial intelligence output that is capable of influencing a covered employment decision, so a resume screening tool that merely ranks candidates for a recruiter’s review falls within the notice duty just as a tool that auto rejects candidates does. The zip code prohibition in subsection (L)(1) is a use restriction rather than a disclosure cure: an employer that discloses a model’s reliance on geographic data has not thereby made that reliance lawful.

The Illinois Department of Human Rights holds rulemaking authority to define the circumstances, timing, and method of notice. IDHR published proposed rules in 2026 but temporarily withdrew them on June 2, 2026, canceling a public hearing scheduled for June 10, 2026, and citing a need for continued coordination with other state agencies without naming a revised timeline. The underlying statutory notice and nondiscrimination duties remain in force notwithstanding the paused rulemaking, which means an employer cannot wait for a final rule before giving notice; it must make a reasonable, good faith judgment about adequate disclosure now. Enforcement proceeds through the standard Illinois Human Rights Act mechanism: a charge filed with IDHR within 300 days of the alleged violation, an IDHR investigation, and, where cause is found, a complaint before the Illinois Human Rights Commission carrying actual damages and attorney’s fees as available remedies.

What happened to the Colorado AI Act and what replaced it

Colorado enacted the first comprehensive state AI statute, Senate Bill 24-205, on May 17, 2024, codified at Colorado Revised Statutes section 6-1-1701 and following. That law defined a high risk artificial intelligence system as one that makes, or is a substantial factor in making, a consequential decision, and defined a substantial factor as AI generated content capable of altering the outcome of a decision affecting employment, credit, housing, healthcare, education, or government services. Like Illinois, Colorado’s original substantial factor test never required the AI system to be the sole or even primary basis for the decision.

The legislature postponed the original law’s effective date twice. SB 25B-004, signed August 28, 2025, during a special legislative session, pushed the start date from February 1, 2026 to June 30, 2026, after Governor Polis warned that the statute in its original form would impose high compliance costs on covered businesses and state and local government. Lawmakers could not reach a compromise on substantive amendments during that special session, so the delay bill did nothing but move the date.

Before the June 30, 2026 date arrived, the legislature replaced the law outright. Governor Polis signed SB 26-189, the Automated Decision-Making Technology Act, on May 14, 2026, repealing the original Colorado AI Act’s discrimination mitigation program and substituting a narrower transparency model that takes effect January 1, 2027. The new law drops the mandatory impact assessment, the risk management program, and the requirement that developers report discriminatory outcomes to the attorney general. In their place, an employer that deploys a covered automated decision-making technology affecting hiring, termination, promotion, compensation, or scheduling must give clear and conspicuous notice before using the technology to materially influence the decision, must disclose within 30 days of an adverse outcome the technology’s role in that outcome in plain language, and must offer the affected individual a right to request meaningful human review where technically feasible, performed by a reviewer who has authority to override the system, considers the relevant evidence, and does not simply defer to the computer generated result. Covered employers must retain compliance records, including version identifiers and changelogs for the technology, for three years. The Colorado Attorney General holds exclusive enforcement authority, and the current Attorney General has stated that the office will not pursue enforcement until the required rulemaking is complete, leaving January 1, 2027 as a formal rather than a practical compliance deadline for now.

What do California’s FEHA automated-decision system regulations require

California’s Civil Rights Council, the body that implements the Fair Employment and Housing Act, adopted automated-decision system regulations that took effect October 1, 2025, amending Title 2 of the California Code of Regulations, including sections 11008 and 11009. The regulations define an automated-decision system as a computational process, derived from artificial intelligence, machine learning, algorithms, or other data processing techniques, that makes a decision or facilitates human decision making regarding an employment benefit such as hiring, promotion, discipline, or termination. By covering a system that merely facilitates a human’s decision, California, like Illinois and Colorado, declines to condition coverage on the AI tool being the sole decision maker.

The regulations do not mandate a bias audit. Instead, section 11008.1 makes evidence that an employer did or did not conduct antibias testing relevant to a discrimination claim and its available defenses, with testing quality measured by its scope, recency, and the employer’s response to adverse results. An employer that never tests its tool for disparate impact forgoes an affirmative defense that a tested, lower risk competitor can raise, which functions as a market incentive toward testing without a statutory mandate to test.

Two further provisions expand exposure for a startup that buys rather than builds its AI hiring tool. Section 11009 extends FEHA’s definition of covered agent to any third party that exercises employment functions, including screening or scoring, on an employer’s behalf, in whole or in part through an automated-decision system, which exposes the vendor itself to direct liability alongside the employer that deployed it. The regulations also extend the recordkeeping period for applications, selection criteria, and automated-decision system data from two years to four years, measured from the later of the record’s creation or the date of the personnel action, giving plaintiffs and the Civil Rights Department a longer evidentiary window than under prior FEHA practice. The regulations do not require a human review step or a candidate appeal mechanism, leaving California’s model as the most purely disclosure and evidentiary regime of the three.

Does the AI tool need to be the sole decision maker for these laws to apply

No state in this group conditions its core obligations on the AI tool acting alone. Illinois’ standard asks only whether an AI output is capable of influencing a covered decision. Colorado’s original substantial factor test asked whether AI generated content was capable of altering the outcome of a consequential decision, and its replacement, SB 26-189, asks whether a covered technology materially influences the decision, again without requiring exclusive control. California’s regulation reaches a system that facilitates human decision making, expressly covering the common case of a tool that ranks or scores candidates for a recruiter who makes the final call.

The practical consequence for a startup is that the common defense of ‘a human always makes the final decision’ does not avoid any of the three states’ notice duties, and in Illinois and under California’s antibias testing incentive it does not reduce discrimination exposure either. The one place AI involvement level does matter is Colorado’s new human review right, which only requires a reviewer where the automated technology materially influenced the outcome and where human review is technically feasible, so a startup that documents a narrow, advisory use of its interview scoring tool has a stronger argument that full human review is already occurring than a startup that auto-rejects candidates below a score threshold.

How should a startup sequence compliance across screening, interview scoring, and performance monitoring

A resume or applicant screening tool triggers the broadest set of obligations because it touches the recruitment and hiring stages named in all three statutes. At minimum, a startup operating in Illinois, Colorado, or California should give applicants notice before the tool is used, confirm with its vendor that no geographic proxy variable feeds the model given Illinois’ zip code prohibition, and request documentation of any disparate impact testing the vendor has run so the employer can decide whether to rely on California’s antibias testing defense.

Interview scoring tools, including video or voice analysis products that generate a candidate score, sit squarely inside Illinois’ notice duty and California’s automated-decision system definition because they facilitate a human interviewer’s decision. Under Colorado’s new law, once SB 26-189 takes effect January 1, 2027, a scoring tool that materially influences an interview outcome will trigger the pre-use notice and, on an adverse result, the 30 day disclosure and human review request right, so a startup should build a documented override path into its interview workflow well before that date rather than retrofitting one under enforcement pressure.

Performance monitoring tools present the least settled exposure of the three use cases. Illinois’ statute reaches the ‘terms, privileges, or conditions of employment,’ which plausibly covers a monitoring score that feeds a discipline or discharge decision, and California’s regulation reaches discipline and termination outcomes directly, so a monitoring tool that informs a performance improvement plan or a termination decision should be treated as covered rather than assumed exempt because it operates after hiring. Colorado’s consequential decision list under SB 26-189 names termination, compensation, and scheduling explicitly, so a scheduling or shift assignment algorithm common in hourly workforce management falls inside the new law’s notice duty even though it never touches the hiring stage.

What does federal litigation add to the compliance picture

Federal case law is supplying the theory of vendor liability that two of the three states have now written into statute. In Mobley v. Workday, Inc., No. 3:23-cv-00770 (N.D. Cal.), Judge Rita Lin ruled on March 6, 2026 that an employer cannot escape Title VII, ADEA, and ADA liability by delegating a traditional hiring function such as resume screening to a third party vendor, holding that Workday can qualify as an employer’s agent for discrimination purposes. The court also held that the ADEA’s protections extend to outside job applicants, rejecting Workday’s narrower reading. The court conditionally certified a collective action covering applicants age 40 and over denied employment recommendations through Workday’s platform from September 24, 2020 forward, with the opt in notice period closing March 7, 2026.

The agent liability theory in Mobley runs parallel to California’s statutory expansion of FEHA’s agent definition to AI vendors and previews the kind of claim Illinois’ IDHR and Colorado’s attorney general are each positioned to pursue against a vendor as well as the employer that deployed its product. A startup’s indemnification and liability allocation language with its AI hiring vendor is now a live legal exposure alongside its own notice practices.

Want more industry insights? I publish daily analysis on the intersection of deep tech and venture law.

Subscribe to The Innovation Attorney on Substack

Open Questions

What will IDHR’s final HB 3773 rules require once rulemaking resumes? Illinois employers are operating under a statutory notice duty with no final regulation defining its content, timing, or delivery method after IDHR withdrew its proposed rules on June 2, 2026. A business leader deciding how much to invest in a compliance program now, rather than after a final rule issues, is pricing in real regulatory uncertainty that a competitor waiting for clarity does not bear.

Will Colorado’s Attorney General complete SB 26-189 rulemaking before January 1, 2027? The Attorney General has stated enforcement will not begin until rulemaking concludes, which means the formal effective date and the practical compliance deadline may diverge. Investors evaluating a Colorado-based or Colorado-serving AI hiring vendor need to know whether the operative deadline is the statute’s date or a later rulemaking date that has not yet been set.

Does California’s voluntary antibias testing defense function as a de facto audit mandate? An employer that forgoes testing loses an affirmative defense a tested competitor can raise, and the four year recordkeeping window gives plaintiffs more time to build a disparate impact case. No published data yet shows whether California’s incentive structure produces testing rates comparable to a hard mandate such as New York City’s Local Law 144 bias audit requirement, a gap with direct cost implications for AI hiring vendors selling into California.

Will the Mobley v. Workday agent liability theory extend into Illinois and Colorado enforcement? Judge Lin’s ruling that a hiring vendor can be a Title VII, ADEA, and ADA agent of the employer supplies a federal common law theory that IDHR and the Colorado Attorney General could invoke independent of the states’ own agent definitions. A startup’s vendor contract allocation of discrimination liability is a live drafting question while this theory is tested on appeal.

How will a multistate employer reconcile Colorado’s human review right with California’s no-review model? Colorado’s SB 26-189 creates a right to request meaningful human review of an adverse decision where technically feasible, while California imposes no equivalent review mandate. A startup building one hiring workflow for both states must decide whether to engineer a human review step as a national baseline or maintain two separate adverse action procedures, a build decision with direct engineering and legal cost.

Source List

  1. 775 ILCS 5/2-102(L) (Illinois Human Rights Act, as amended by H.B. 3773, 2024). Illinois General Assembly, enacted 2024, effective January 1, 2026. Primary statutory source for the notice duty and the discriminatory use and zip code proxy prohibitions discussed throughout the Illinois findings.
  2. Colo. Rev. Stat. sec. 6-1-1701 et seq. (Colorado Artificial Intelligence Act, S.B. 24-205). Colorado General Assembly, signed May 17, 2024. Primary statutory source for the original substantial factor and consequential decision definitions, since repealed and replaced.
  3. S.B. 26-189, Automated Decision-Making Technology Act (Colo. 2026). Colorado General Assembly, signed May 14, 2026, effective January 1, 2027. Primary statutory source for Colorado’s current notice, adverse action disclosure, human review, and recordkeeping duties.
  4. 2 Cal. Code Regs. secs. 11008, 11008.1, 11009. California Civil Rights Council, effective October 1, 2025. Primary regulatory source for California’s automated-decision system definition, antibias testing defense, agent liability, and recordkeeping extension.
  5. Mobley v. Workday, Inc., No. 3:23-cv-00770-RFL (N.D. Cal. Mar. 6, 2026 order). U.S. District Court for the Northern District of California, March 6, 2026. Primary case authority for the agent liability theory extending Title VII, ADEA, and ADA exposure to AI hiring vendors.
  6. Seyfarth Shaw LLP, “Illinois Department of Human Rights Temporarily Withdraws Proposed Rules on Use of Artificial Intelligence in Employment”. Published June 8, 2026. Reports the June 2, 2026 withdrawal of IDHR’s proposed HB 3773 rules and confirms the statutory duty remains in force.
  7. Akin Gump Strauss Hauer & Feld LLP, “Colorado Postpones Implementation of Colorado AI Act, SB 24-205”. AI Law and Regulation Tracker, published 2025. Confirms the terms and rationale of SB 25B-004’s postponement of the original Colorado AI Act to June 30, 2026.
  8. Buchalter, “Colorado Rewrites Its AI Law: What Employers Must Know About SB 26-189”. Published 2026. Secondary source detailing SB 26-189’s definitions, notice timing, human review standard, and recordkeeping period.
  9. Mayer Brown LLP, “California Adopts New Employment AI Regulations Effective October 1, 2025”. Published August 2025. Secondary source summarizing the California Civil Rights Council regulations’ citation, definitions, and liability provisions.
  10. Littler Mendelson P.C., “Colorado Amends its Artificial Intelligence Law, Substantially Reducing Obligations on Employers”. Published 2026. Secondary source comparing the original Colorado AI Act’s obligations to SB 26-189’s narrower replacement structure.
  11. The Innovation Attorney Governance for HR, “The Mobley v. Workday Case Didn’t End. It Escalated. Here’s Where It Stands.”. Published May 2026. Secondary source providing the procedural posture, case number, and class certification status of Mobley v. Workday as of May 2026.

Bibliography

  • 775 Ill. Comp. Stat. 5/2-102(L) (2026).
  • Colo. Rev. Stat. sec. 6-1-1701 to -1707 (2024) (repealed 2026).
  • S.B. 26-189, 75th Gen. Assemb., Reg. Sess. (Colo. 2026) (codified as amended at Colo. Rev. Stat. sec. 6-1-1701 et seq., eff. Jan. 1, 2027).
  • S.B. 25B-004, 75th Gen. Assemb., 2d Spec. Sess. (Colo. 2025).
  • 2 Cal. Code Regs. secs. 11008, 11008.1, 11009 (2025).
  • Mobley v. Workday, Inc., No. 3:23-cv-00770-RFL (N.D. Cal. Mar. 6, 2026) (order on motion to dismiss and conditional certification).
  • Akin Gump Strauss Hauer & Feld LLP. “Colorado Postpones Implementation of Colorado AI Act, SB 24-205.” AI Law and Regulation Tracker, 2025.
  • Buchalter. “Colorado Rewrites Its AI Law: What Employers Must Know About SB 26-189.” Buchalter Insights, 2026.
  • Mayer Brown LLP. “California Adopts New Employment AI Regulations Effective October 1, 2025.” Mayer Brown Legal Update, August 2025.
  • Littler Mendelson P.C. “Colorado Amends its Artificial Intelligence Law, Substantially Reducing Obligations on Employers.” Littler ASAP, 2026.
  • Seyfarth Shaw LLP. “Illinois Department of Human Rights Temporarily Withdraws Proposed Rules on Use of Artificial Intelligence in Employment.” Seyfarth Legal Update, June 8, 2026.
  • The Innovation Attorney Governance for HR. “The Mobley v. Workday Case Didn’t End. It Escalated. Here’s Where It Stands.” May 2026.

Interested in analysis about the intersection of tech, policy and the law? Check out my Substack channel. https://theinnovationattorney.substack.com/