Illinois HB 3773: What Employers Must Do When AI Touches a Hiring Decision

A recruiter in Chicago opens the ATS on a Monday morning and the shortlist is already sorted. The tool scored 180 applications overnight. Nobody on the team wrote the word "algorithm" in the job ad. Under Illinois law, that quiet overnight sort can already be a problem, and it has been one since 1 January 2026.

House Bill 3773, enacted as Public Act 103-0804 and signed on 9 August 2024, amended the Illinois Human Rights Act. The operative text sits at 775 ILCS 5/2-102(L). It does two things that employers keep collapsing into one. Using artificial intelligence in a way that has a discriminatory effect is a civil rights violation. Failing to tell people you are using it for a covered employment decision is a separate civil rights violation. The second one does not depend on proving uneven outcomes.

The Illinois Department of Human Rights was told to write the notice rules. It published a proposal on 15 May 2026 and withdrew it on 2 June 2026, citing a need to keep working with other state agencies. No refiling date has been published. The statute did not pause with the proposal.

What the statute actually prohibits

Paragraph (L)(1) makes it a civil rights violation to use AI that has the effect of subjecting employees or applicants to discrimination on the basis of a protected class. Intent is not the test. A vendor that describes the model as neutral does not move the employer outside the paragraph.

The covered decisions track the rest of the employment article: recruitment, hiring, promotion, renewal of employment, selection for training or apprenticeship, discharge, discipline, tenure, and the terms, privileges, or conditions of employment. Screening is inside that list. So is a promotion model, a training-nomination tool, and a performance score that feeds a termination.

The same paragraph bars the use of zip code as a proxy for a protected class. Ordinary location data is not banned. Using a postal code, or a feature derived from one, as a stand-in for race, national origin, or another protected characteristic is. Ask the vendor in writing whether zip code, postal geography, or a derived geographic feature enters the model. A notice does not cure a proxy.

Paragraph (L)(2) is the notice duty. Failing to provide notice that the employer is using AI for those same purposes is itself a civil rights violation. A model with even selection rates still fails this paragraph if nobody was told it was running.

Enforcement uses the machinery the Act already has. A person files a charge with IDHR. The case can move to the Human Rights Commission. Remedies include damages, injunctive relief, and attorney's fees. There is no new registration portal and no annual filing, which is the main practical difference from New York City Local Law 144.

Who is covered, and which tools count

Since Public Act 101-0430 took effect on 1 July 2020, the Illinois Human Rights Act reaches employers with one or more employees in Illinois during 20 or more calendar weeks in the calendar year of the alleged violation or the year before. There is no 15-employee threshold on this amendment. A Texas company that employs people in Illinois for that period is in scope for those Illinois employees. Public employers, hospitals, and universities are not carved out.

Section 2-101 defines artificial intelligence as a machine-based system that, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments. The definition includes generative AI. It does not require a neural network, and it does not require the system to be the sole basis for the decision. The output has to be capable of influencing one.

That reaches more than the product marketed as an "AI screener":

An inventory that only lists tools the team calls "AI" will miss the match score buried in the ATS settings. The statutory test is what the system does, not the label on the invoice.

Illinois already had a narrower statute on AI analysis of video interviews, the Artificial Intelligence Video Interview Act. HB 3773 does not replace it. A video score can trigger both: consent and explanation duties under the video law, and the effect-plus-notice duties under the Human Rights Act.

The rules that are not in force

HB 3773 told IDHR to adopt rules on when notice is required, how long the employer has to give it, and how to deliver it. The rulemaking has stalled.

In December 2025 the Department shared an informal draft, Subpart J, with stakeholders. That draft was never formally published, so it had no procedural status. On 15 May 2026 IDHR published proposed amendments to Part 2520 of the Illinois Administrative Code (Title 56, Labor and Employment) in the Illinois Register. The comment period was due to close on 29 June 2026, with a hearing set for 10 June. On 2 June the Department postponed the rulemaking, cancelled the hearing, and said it needed more time to work with other agencies. As of early October 2026 the proposal has not been refiled.

The withdrawn draft is not a compliance standard. It is still the clearest written signal of how the enforcing agency was reading the statute. Treating it as a design input, and saying so in your file, is a reasonable posture. Treating a sentence from it as adopted law is not.

As proposed, the draft would have required notice whenever AI output influences or facilitates a covered employment decision. Targeted recruitment advertising and resume screening were named. The notice elements on the table included the developer, product name, and vendor; the decision affected; the purpose; the categories of information processed; the job positions; a contact for questions; and information on requesting a reasonable accommodation. Timing in the draft was annual notice to current employees, notice within 30 days of adopting a new or substantially updated system, and notice to applicants in the job posting. Delivery was layered: handbook or policy, workplace posting, intranet or external site, and the posting itself.

Two details in secondary summaries of that draft do not line up, which is a reason not to build a calendar on them. Retention figures in published alerts disagree, with some saying three years and some saying four, and at least one reading of the formal proposal finds no new retention number at all. The statute's notice paragraph is also phrased around employees, while the discrimination paragraph expressly includes applicants. The draft closed that gap by requiring notice in the job posting. With the draft withdrawn, the textual gap is open again. Notifying applicants anyway is the low-cost reading, because the agency already wrote that it expected posting notice, and because over-notifying does not create a violation.

How this differs from NYC, Colorado, and California

Teams that already built a Local Law 144 file sometimes assume Illinois is the same packet with a different letterhead. It is not.

California added another layer on 30 September 2026, when the governor signed SB 947, the No Robo Bosses Act. That law, effective 1 July 2027, limits sole reliance on an automated decision system for discipline and termination. It does not change the Illinois notice duty. A multi-state employer needs a matrix, not one national paragraph pasted into every posting.

For the wider map, including the EU AI Act's high-risk classification for recruitment systems, see AI hiring compliance in 2026.

A workflow that holds up without final rules

"Compliant" is the wrong word until IDHR refiles. "Documented and reasonable" is the standard an investigator can actually look at.

1. Inventory by function. Walk recruitment, hiring, promotion, training selection, discipline, and discharge. Record every system whose output influences a step, including modules you did not buy as a separate product.
2. Get the disclosure facts from the vendor in writing: product name, developer, purpose, data categories, and whether zip code or a geographic feature is used. If the contract forbids you from naming the vendor in a candidate notice, that is a procurement problem, not a reason to skip notice.
3. Publish a short notice in the job posting and in the employee handbook or policy. Plain language. A contact for questions and for accommodation requests. Languages the workforce actually uses, and a format a screen reader can reach. The draft asked for both, and agencies rarely drop accessibility between drafts.
4. Refresh the employee notice on a set date each year, and again within 30 days of a new tool or a material change to scoring. Those intervals came from the withdrawn draft. Using them is a choice you can explain.
5. Keep the notices, the postings, the version, the date, and how you distributed them. Because the retention figure is unsettled, keep them at least as long as your longest other applicant-record schedule. California's four-year ADS retention is a useful outer marker if you hire there too. See how long to keep candidate data.
6. Test for effect, even though Illinois did not order an audit. The prohibition is effects-based. A quarterly comparison of selection rates, using the four-fifths rule as a screen rather than a verdict, is the minimum that makes the (L)(1) file real. The AI hiring audit checklist is a workable sequence. The mechanisms worth testing are the same ones in bias in AI hiring: training history, proxy variables, and cutoffs that turn a small score gap into a rejection.
7. Watch the docket. A refiled Subpart J will open another comment window. A program that already exists is easier to adjust than one you start the week the rules return.

What to put in the posting

A notice that only says "we may use AI" is thin. The withdrawn draft asked for specifics, and a charge investigator will ask for them whether or not those specifics are codified. A posting paragraph can be short:

We use [product name], from [vendor], to score applications for this role against the requirements in the job description. The score influences who is invited to a first interview. It uses the resume and the answers on this form. It does not use zip code. Questions, including requests for a reasonable accommodation: [email protected].

That paragraph is an example, not a statutory template. Adjust it when the tool also reads a cover letter, a work sample, or a video. If a human reads every application before anyone is rejected, say that, because it is a fact about the decision, not a slogan.

Candidates are also writing with models. Illinois does not require applicants to disclose that. If you want a disclosure, keep it separate from the employer notice, and read what to do with candidate AI disclosure before you add a checkbox that you cannot score fairly.

Where screening software fits

The law does not ban scoring. It bans a discriminatory effect, a zip-code proxy, and silence. A screening setup that shows the requirement behind a score is easier to defend than a single match percentage with no breakdown, because you can see whether the model is reacting to a real job requirement or to something you would not put in a posting. Resume World scores each application against the rubric you set and shows the reason for the score, which is the record you want if someone later asks why a candidate was not advanced. The legal duty stays with the employer. A vendor explanation does not transfer it.

If you hire in New York City as well, the Illinois notice does not satisfy the Local Law 144 audit. If you hire in California, the Illinois file does not replace the four-year ADS retention rule. Run the Illinois steps for Illinois roles, and keep the other states on their own rows.

More on hiring rules and screening

• AI Hiring Compliance in 2026: What Changed and What to Do
• NYC Local Law 144: A Practical Guide for Employers
• Colorado SB 26-189: What Employers Need to Do
• Bias in AI Hiring: The Three Mechanisms, and How to Test for Them
• The Four-Fifths Rule in Hiring: How to Use It Without Misreading It
• AI Hiring Audit Checklist
• Candidate Data Retention: How Long to Keep Applications
• When Candidates Disclose They Used AI