Skip to content

Illinois HB 3773: AI in Employment Decisions from 2026

9 min read · updated August 11, 2026

HB 3773 adds no audit, no registration and no new agency. It writes AI into a civil rights statute Illinois has had since 1979, and that choice determines everything about how it will be enforced.

An amendment to the Human Rights Act

House Bill 3773 was signed on 9 August 2024 as Public Act 103-0804 and took effect on 1 January 2026. It amends the Illinois Human Rights Act, 775 ILCS 5, adding a definition of artificial intelligence to section 1-103 and new employer prohibitions in Article 2. The definition is deliberately broad — a machine-based system that infers from input how to generate outputs that can influence physical or virtual environments — and it expressly includes generative AI. There is no threshold, no exemption for small employers beyond the Act’s existing coverage rules, and no carve-out for off-the-shelf tools.

Because it is an amendment, the surrounding machinery comes along unchanged. The covered employment actions are the ones the Human Rights Act already lists: recruitment, hiring, promotion, renewal of employment, selection for training or apprenticeship, discharge, discipline, tenure, and the terms, privileges or conditions of employment. The protected classes are the Act’s, which are broader than Title VII’s — Illinois includes, among others, order of protection status, military status, and work authorisation status. The public act text is at the Illinois General Assembly.

Not legal advice. Whether a given tool touches a covered employment action, and whether your organisation is a covered employer under 775 ILCS 5/2-101, are threshold questions to put to employment counsel before anything else on this page matters.

The two prohibitions

The amendment makes two things civil rights violations, and they are different in kind.

The first is effects-based. An employer may not use artificial intelligence that has the effect of subjecting employees to discrimination on the basis of a protected class with respect to a covered employment action. Note the phrase: has the effect of. There is no intent element. A vendor tool that the employer neither built nor understands, producing a disparity the employer did not want, is within the prohibition on its face. This is a codification of disparate impact into the AI context, and it is the opposite drafting choice from Texas TRAIGA, which requires intent and says expressly that disparate impact alone does not establish it.

The second is a specific proxy ban. An employer may not use zip code as a proxy for a protected class. Zip code is named because American residential geography encodes race with unusual fidelity, and because it is the feature most likely to survive a naive “we removed race from the model” exercise. The ban is narrow in its terms — it names one variable — but the effects-based prohibition above it does the general work, so removing zip code and keeping a distance-to-office feature that reproduces it is not a defence.

The notice duty, and what is still unwritten

The amendment also requires an employer to provide notice when it uses artificial intelligence for a covered employment action, and directs the Illinois Department of Human Rights to adopt rules on the circumstances, conditions, time and means of that notice.

The statute itself does not say when the notice must be given, what it must contain, or in what form. That is the part that has to come from IDHR rulemaking, and until those rules are final an employer designing a notice is working from the statutory sentence and analogy. The nearest available analogies are NYC Local Law 144’s ten-business-day advance notice and the Illinois AI Video Interview Act’s pre-interview explanation duty, neither of which binds here. Check the current status of IDHR rulemaking before treating any notice template as adequate.

An obligation whose content is set by rules that may not yet be final is a genuinely unsettled compliance question. The prudent position is notice that is early, specific about the fact of AI use and the decision it informs, and documented — but that is a risk judgement, not a statement of what the rules require.

How a claim actually proceeds

There is no new cause of action and no direct route to court. A claim runs through the Human Rights Act’s existing process: the complainant files a charge with the Illinois Department of Human Rights, generally within 300 days of the alleged violation; IDHR investigates and issues a determination; and on a finding of substantial evidence, or on the complainant’s election, the matter proceeds either to the Illinois Human Rights Commission or to circuit court. Remedies are the Act’s: back pay, hiring or reinstatement, damages for emotional distress, attorney’s fees, and injunctive relief.

The consequence worth internalising is evidentiary. An effects-based prohibition is proved with data — selection rates by protected class across the population the tool touched. An employer who has never computed that has no way to answer the charge, while one who computes it quarterly has both an early warning and a record. That is the same practical logic that makes the New York City bias audit worth doing even where it is not required.

Federal law runs in parallel and is not displaced. Title VII, the ADEA and the ADA apply to the same conduct, and the EEOC’s technical assistance on selection procedures and adverse impact remains the reference point for how the analysis is done — see the EEOC guidance page. An Illinois employer can be inside HB 3773 and inside Title VII on one set of facts, with different limitation periods and different remedies.

Why this is not a bias-audit law

It is worth being explicit, because the three American employment-AI instruments are routinely conflated. New York City requires an independent bias audit, published annually, and fines you for not having one — but the audit itself creates no liability for the disparity it reveals. Colorado’s AI Act imposes a duty of reasonable care against algorithmic discrimination together with impact assessments, and gives enforcement to the Attorney General alone. Illinois does neither: it requires no audit and no assessment, and instead makes the discriminatory effect itself a civil rights violation actionable by the affected individual.

That is a materially stronger position for a rejected candidate than either of the others, and a materially thinner compliance checklist for the employer. Nothing here can be discharged by producing a document. The only defence is that the tool did not produce the effect, which is a question about your data — which is why an employer with no measurement has, in practical terms, no defence.