Illinois's AI Employment Rule Starts With a Three-Part Distinction
Illinois's Public Act 103-0804 changes the Illinois Human Rights Act by adding definitions for artificial intelligence and generative artificial intelligence, then adding a new provision on the use of artificial intelligence in employment. The important question is not simply whether the state has decided that “biased AI” is unlawful. It is how the amendment separates three different legal concerns: what an AI-assisted employment practice does, whether zip codes are being used as a proxy for a protected class, and whether the employer tells an employee that covered AI is being used. Understanding that structure gives employers, workers, and other readers a more reliable way to judge an automated employment process without confusing one form of accountability with another.
A focused amendment with consequences beyond hiring software
The law is Public Act 103-0804, an enrolled House Bill 3773. It amends Sections 2-101 and 2-102 of the Illinois Human Rights Act, and its effective-date provision says that the Act takes effect on January 1, 2026. That matters because the amendment is not a voluntary principle or a proposal waiting for a later decision. It places AI-related language inside an existing civil-rights statute governing employment.
The placement does much of the legal work. An employer may use a vendor's screening tool, a scoring model, a recommendation engine, or a generative system rather than building software internally. But changing the technology provider does not change the employment relationship in which the system is used. If the tool ranks candidates, recommends whom to promote, helps determine access to training, or influences discipline, the relevant question becomes what the employer's use does within a protected employment context. The amendment makes that use visible to the Illinois Human Rights Act rather than leaving it in a separate category called “technology.”
The Illinois Department of Human Rights presents the change in similar, but less legally precise, terms. On its legislative-updates page, the Department describes recent legislation on artificial intelligence in employment as part of Illinois's response to emerging forms of discrimination and notes that technology can perpetuate bias. That is useful context, but it is agency explanation, not a substitute for the enacted text. The Public Act supplies the operative rule. The Department's page helps show why the subject is being treated as a civil-rights issue and identifies employers, legal practitioners, advocates, and affected communities as relevant audiences.
The amendment's reach is also not confined to large technology companies. The Act's employer definition includes a person employing one or more employees in Illinois during 20 or more calendar weeks in the relevant year or preceding year. It separately includes the State and political subdivisions, parties to public contracts, and joint apprenticeship or training committees without regard to employee count. The Department summarizes the broader point by saying that the Illinois Human Rights Act applies to most employers regardless of size. The exact statutory categories and exclusions still matter, but a small organization should not assume that a small budget or a purchased tool places it outside the law.
This creates a practical tension. Automation can make a decision look less personal and more objective: a score appears to come from data, a recommendation appears to come from a model, and a vendor may describe the system in neutral business language. Civil-rights law asks a different question. It asks whether the employer's use affects a protected employment decision in a way the statute addresses. The source of the score does not, by itself, answer that question.
One provision, three different questions
Section 2-102(L), titled “Use of artificial intelligence,” contains the core architecture. Paragraph (1) addresses the substantive use of AI. In the listed employment contexts, it prohibits an employer from using artificial intelligence that has the effect of subjecting employees to discrimination on the basis of protected classes under the Article. In the same paragraph, the text separately names the use of zip codes as a proxy for protected classes under the Article.
Paragraph (2) addresses notice. It makes it a civil-rights violation for an employer to fail to provide notice to an employee that the employer is using artificial intelligence for the purposes described in paragraph (1). The Department must adopt rules for implementation and enforcement, including rules about the circumstances and conditions that require notice, the time period for providing it, and the means of providing it.
Those sentences create three questions that can arise from one employment process.
First: what effect did the practice have? Suppose an employer uses an automated ranking system during recruitment. The system's stated objective might be to identify applicants likely to succeed in a role. The effects inquiry looks past that objective to the consequences of the employer's use: did the practice subject employees to discrimination on a protected basis? The statutory wording centers the effect. It does not state, in this provision, a numerical disparity threshold, a required comparison group, or an additional requirement that the employer intended discrimination. That does not mean every unequal result establishes a violation; it means the supplied text gives an effects-focused rule without supplying the complete method for proving it.
Second: what role did geography play? If the system uses a zip code, the question is not limited to the final distribution of scores. The statute expressly names using zip codes as a proxy for protected classes. A zip code might enter as a model feature, an eligibility filter, a weighting factor, or an instruction used by a person who relies on the system. The proxy question follows that input path. It asks what the geographic variable is doing and whether it stands in for a protected class, rather than treating geography as an incidental data field.
Third: was notice provided? Even if an employer believes the system has neutral effects and uses no improper geographic proxy, the employer still has to consider the separate notice language when the AI is used for the covered purposes. Notice is not described as a cure for discriminatory effects. Nor does notice answer whether zip code was used as a proxy. It is a transparency obligation with its own unresolved questions about trigger, timing, content, and delivery.
Consider a deliberately simple example. An employer uses an AI-assisted hiring screen. The tool produces a candidate ranking; a zip code is one of its inputs; and the employer does not tell an employee that AI is being used. A reviewer should not reduce the situation to one question—“Was the model biased?” They should keep at least three records of inquiry: the result across relevant protected groups, the path by which the zip code entered the process, and the notice event. A different outcome could change one inquiry without eliminating the others. A system might show no measured disparity in a small sample while still raising a question about the role of zip code. Or an employer might provide notice while the resulting practice still has a discriminatory effect. The amendment's structure keeps those possibilities analytically separate rather than collapsing them into one overall label.
This is a reasoning model, not a statutory flowchart that decides liability. Its value is that it prevents a conclusion on one branch from being silently reused on another. An employer's notice record is evidence about notice; it is not automatically evidence that the AI's outcome was nondiscriminatory. A disparity analysis is evidence about effects; it does not by itself explain whether the employer used zip code as a proxy. And finding a geographic proxy does not make the notice question irrelevant.
The law is clear about the separation, but not complete about the tests
The amendment gives readers a firm starting point and a firm reason for caution. The starting point is the separation of duties. The caution comes from the details the cited text leaves for the Department or later interpretation.
For the effects pathway, the Act does not state how much disparity is enough, what population should be compared, how causation is shown, or what defenses may apply. For the proxy pathway, it does not define “proxy,” identify particular zip codes, or explain whether a measurable adverse outcome or an intent to discriminate must be shown first. The text places the proxy language within the AI-use subsection, but the sentence does not repeat the words “use artificial intelligence” immediately before the proxy phrase. Its exact scope therefore deserves careful interpretation rather than an assumption that every geographic use, or only an AI-mediated geographic use, has already been fully defined.
Notice has a different kind of incompleteness. The duty itself is written into the statute, but the Department must specify the circumstances and conditions that require notice, the time period for giving it, and the means of giving it. The Department's legislative page says that proposed administrative rules implementing the Illinois Human Rights Act typically have 45-day public-comment periods and are published in the Illinois Register. That explains how implementation may develop; it does not establish that a particular AI notice rule has been proposed or finalized. Until the relevant rules and interpretations are available, a reader should not manufacture exact notice wording or timing from general assumptions about transparency.
Nor should AI-use notice be confused with the Act's general workplace notice. Section 2-102(K) concerns a Department-prepared or Department-approved notice summarizing rights under the Article and information about filing a charge. It can be posted where workplace notices are customarily posted or included in an employee handbook. That provision also describes a procedure under which the Department may investigate a reported failure, give an employer 30 days to correct it, and potentially initiate a charge if the violation is not corrected. Section 2-102(L)(2) instead speaks to telling an employee that the employer is using artificial intelligence for the purposes in paragraph (1), with the details left to AI-specific rules. A handbook update may be useful, but the two provisions should not be presumed interchangeable.
This distinction is the reason to resist two opposite mistakes. The first is complacency: an employer discloses an AI tool and treats disclosure as proof that the process is lawful. The second is overstatement: a reader sees the words “AI,” “proxy,” and “notice” and assumes the Act already supplies a complete audit protocol, a fairness metric, a vendor-liability rule, or a safe harbor for human review. The selected statutory text supplies none of those technical specifications. It supplies a legal architecture within which facts, agency rules, and later interpretation will matter.
The most useful conclusion at this point is therefore modest but concrete. Illinois has not made AI employment analysis a single question about whether a model is good or bad. It has placed three accountability questions beside one another: what the employer's AI-assisted practice does to people in covered employment settings, whether zip code is being used as a protected-class proxy, and whether the affected employee receives the required notice. To apply that architecture responsibly, the next step is to determine what counts as AI under the Act, which employer and worker relationships are within its definitions, and which employment decisions the text names. Scope must come before judgment.