Illinois HB 3773: The AI Hiring Law With No Bias-Audit Safe Harbor
Most employers built their AI hiring compliance program around New York City's model: commission an audit, publish a summary, post a notice, move on. Illinois wrote a different kind of law. Since January 1, 2026, AI in hiring sits inside the Illinois Human Rights Act itself — where the discriminatory outcome is the violation and there is nothing to publish your way out of.
Illinois Did Not Write a Procedural Law
The reason HB 3773 keeps catching employers is that it looks, from a headline, like the third entry in a familiar list — New York City, then Colorado, then Illinois. It is not. NYC Local Law 144 is a disclosure regime with a defined path to compliance: get the independent bias audit, publish the summary of results, notify candidates ten business days ahead, and the automated employment decision tool may lawfully be used. Do the steps, satisfy the law.
Illinois amended its core civil rights statute instead. HB 3773 inserted artificial intelligence directly into the Illinois Human Rights Act's definition of what an employer may not do. The operative question under Illinois law is not "did you complete the required audit steps" but "did the tool have the effect of discriminating against a protected class." That is a substantive standard, and it does not come with a checklist that ends in immunity.
What the Statute Reaches
Far More Than Resume Screening
SCOPE TRAPThe covered decision types run across the employment lifecycle — recruitment, hiring, promotion, renewal of employment, selection for training or apprenticeship, discharge, discipline, tenure and other terms, privileges or conditions of employment. Employers who scoped their review to the applicant-tracking system missed the AI in their performance-management, scheduling and internal-mobility tooling.
Discriminatory Effect, Not Discriminatory Intent
EFFECT STANDARDThe violation is defined by the effect of subjecting employees to discrimination on the basis of a protected class. An employer who never intended a disparate outcome, and whose vendor never inputs a protected characteristic, is still exposed if the outcomes break along protected lines.
Zip Code Singled Out as a Proxy
NAMED IN STATUTEThe statute separately prohibits using zip code as a proxy for a protected class. Illinois legislators watched the proxy-variable problem play out in lending and named the most common one directly — which means an employer cannot argue the variable was facially neutral and therefore permissible.
An Affirmative Notice Duty
SEPARATE OBLIGATIONEmployers must notify employees when AI is used for the covered decision types, with the Illinois Department of Human Rights empowered to set rules on timing and manner. Notice is a separate obligation from the anti-discrimination provision — an employer can produce clean outcome data and still have a notice failure.
Enforcement Through the Existing Civil Rights Machinery
ENFORCEMENT PATHBecause the AI provisions live in the Human Rights Act, they are enforced the way every other Illinois discrimination claim is — a charge with the Department of Human Rights, investigation, and then the Human Rights Commission or circuit court. There is no new AI-specific fine schedule to budget against, which employers misread as leniency rather than as ordinary discrimination exposure.
Why a NYC-Shaped Program Fails an Illinois Test
A Local Law 144 program is built to produce two artifacts: an audit summary on a public web page, and a candidate notice. Both are point-in-time. The audit is typically annual, covering the tool as configured on the day it was measured, and the published impact ratios describe the vendor's aggregate data as often as they describe the employer's own hiring. That package answers the question New York City asks.
Illinois asks whether this employer's use of the tool produced a discriminatory effect on this workforce. A vendor-level audit ratio is thin evidence of that. What actually answers the Illinois question is the employer's own outcome data — selection rates by protected class at each automated stage of its own funnel, tracked continuously, with a documented record of what was changed when a gap appeared. Employers who have the NYC artifact and nothing else have documentation that was designed for a different question.
Building an HB 3773 Compliance Program
Inventory AI across the whole employment lifecycle, not just hiring
Build a register of every tool that scores, ranks, routes, flags or recommends a person — including performance analytics, scheduling optimizers, internal mobility matching and attrition-risk models. The statute's covered decisions extend to discipline, discharge and terms of employment, so a hiring-only inventory understates the exposure.
Measure selection rates on your own funnel, continuously
Track pass-through rates by protected class at each automated stage using your own applicant and employee data rather than relying on the vendor's aggregate impact ratios. Illinois liability attaches to your outcomes, so your data is the only evidence that speaks to the actual legal question.
Hunt proxies beyond the one the statute names
Zip code is called out explicitly, but the same correlation problem runs through commute distance, high school or university, employment-gap penalties, name-derived features and social or device signals. Test whether removing a suspect variable materially changes outcome distributions, and document what you found either way.
Make notice a per-process, timestamped record
Attach the AI disclosure to the specific process where the tool runs — the job posting, the assessment invitation, the performance-review workflow — and log when it was shown and to whom. A generic privacy-policy paragraph is difficult to offer as proof that a particular employee received notice for a particular decision.
Preserve a human decision record with independent reasons
Where a human reviews the model's output, capture the reasons that reviewer actually relied on. A human who reliably ratifies the ranking is not meaningful oversight, and an outcome record showing the score was always followed makes the tool, not the reviewer, the effective decision-maker.
Negotiate vendor terms as a financial hedge, not a defense
Push for indemnity, the right to receive underlying scoring data, cooperation in a Department of Human Rights investigation, and notice of material model changes. Understand these recover money and evidence — they do not move the statutory obligation, which stays with the employer that made the decision.
The Remote-Workforce Problem
The employers most likely to be caught flat by HB 3773 are the ones with no Illinois office. The Human Rights Act reaches employment in Illinois, and a distributed company hiring for remote roles routinely ends up with Illinois-based employees and Illinois-based applicants running through the same automated funnel as everyone else. A compliance scope defined by office locations rather than by where candidates and employees actually sit will miss this entirely — and the notice obligation, being affirmative, generates a violation through pure inaction rather than through any bad outcome.
Frequently Asked Questions
We already comply with NYC Local Law 144. Does that cover Illinois?
It covers part of the notice practice and gives you a starting point on measurement, but it is not sufficient. Local Law 144 compliance produces a point-in-time, often vendor-level audit designed to answer a procedural question. Illinois asks a substantive one about the effect of your use of the tool on your workforce, and there is no Illinois provision that makes an audited tool lawful to use. Treat the NYC artifact as an input, not as the program.
Does HB 3773 apply to AI used only to source or advertise roles, before anyone applies?
Recruitment is inside the covered decision list, so sourcing and audience-targeting tools are in scope rather than outside it. Ad-delivery optimization that concentrates a job's audience along age, sex or neighborhood lines is exactly the kind of upstream effect the statute's language reaches, and it is a common blind spot because the exposure sits with the media platform's model rather than with the ATS.
What if our AI tool only ranks candidates and a recruiter makes the final call?
Human-in-the-loop reduces risk only to the extent the human genuinely changes outcomes. If your data shows recruiters advance the top-ranked candidates at close to a hundred percent and rarely reach past a certain rank, the ranking is functionally the decision, and the effect of that ranking on protected classes is what the statute measures. The defensible version is a documented review record showing independent reasons and real deviation from the model's order.
Is there a penalty schedule specific to HB 3773?
No, and that is the point employers misread. Because the provisions were folded into the Human Rights Act rather than enacted as a standalone AI statute with its own fines, an AI-driven violation carries the same remedies as any other Illinois civil rights violation — which can include back pay, reinstatement, damages, attorney's fees and injunctive relief, plus the Department of Human Rights investigation process itself.
How does this interact with the Illinois AI Video Interview Act?
They are separate obligations that stack. The AI Video Interview Act's consent, explanation, deletion and demographic-reporting duties for AI-analyzed video interviews continue to apply on their own terms, and HB 3773 adds Human Rights Act exposure on top for the same tool. Employers running AI video interviews in Illinois need both programs, not a merged one.
What should we do first if we have done nothing yet?
Start with the inventory and the notice, in that order, because those are the two failures that are both easy to prove and entirely within your control. You cannot retroactively fix a discriminatory outcome from last quarter, but you can stop accruing notice violations this week — and the inventory is what tells you how large the outcome-measurement problem actually is.
Related AI hiring compliance guides
Find AI Hiring Audit and Governance Tools on RatedWithAI
RatedWithAI reviews AI governance and bias-testing platforms — including tools built for adverse-impact measurement, model inventories and the outcome documentation Illinois employers now need to keep.
Explore AI Legal & Compliance Guides