Illinois regulates AI in employment through two distinct statutes: an amendment to the Illinois Human Rights Act covering AI-assisted employment decisions generally, and the longer-standing Artificial Intelligence Video Interview Act covering AI analysis of recorded interviews specifically. Neither law dictates a particular technical implementation — both are satisfied by notice to the employee or candidate, a documented non-discrimination check, and a human decision-maker attached to the outcome. This post covers what each statute asks for and how a hiring agent should be built to satisfy both.

For the broader state AI law landscape this sits in, see US state AI laws compared.

Two separate statutes, two separate triggers

It is easy to conflate Illinois's AI employment rules into one law. They are not the same statute, and they trigger on different events.

The Human Rights Act amendment applies broadly to the use of AI in employment decisions — recruitment, hiring, promotion, discipline, and termination. Its core duties are notice to employees that AI is being used in a covered decision, and a prohibition on using AI in a way that results in unlawful discrimination based on a protected characteristic, including through the use of a proxy — the statute specifically calls out zip code as an example of a variable that can function as a discriminatory proxy for race or national origin even when it is not explicitly a protected category.

The Artificial Intelligence Video Interview Act is narrower and older: it applies specifically when an employer uses AI to analyze video interviews of applicants for positions based in Illinois. It requires notifying the applicant before the interview that AI will be used, explaining in general terms how the AI works and what characteristics it evaluates, and obtaining the applicant's consent before recording. It also restricts sharing the video and requires deletion within a defined window after a candidate requests it.

A hiring agent that only screens resumes and never touches video is covered by the first statute and not the second. An agent that scores recorded interview responses is very likely covered by both.

What "AI in employment decisions" covers for an agent

The Human Rights Act amendment is written broadly enough to cover most of what a modern hiring agent does:

  • Resume screening and ranking — an agent that filters or scores candidates before a human sees them.
  • Interview scheduling and pre-screening chat — a conversational agent that asks qualifying questions and passes a recommendation forward.
  • Promotion and performance-review support — an agent that surfaces performance data or a recommendation feeding into a promotion decision.
  • Discipline and termination support — an agent whose output influences a disciplinary or termination outcome.

The common thread is not "is a human technically in the loop" but "does the AI's output materially shape the outcome." An agent that produces a ranked shortlist a recruiter almost always follows is functionally driving the decision, even if a human clicks the final button.

The proxy-discrimination problem, specifically

Regulators calling out zip code by name is a signal worth taking literally. Geographic variables correlate strongly with race and national origin in many US metro areas, and a hiring agent that uses location — directly, or indirectly through a feature like "distance to office" or "current employer industry cluster" — can reproduce discriminatory outcomes without anyone writing a discriminatory rule.

The engineering response is not "remove zip code from the input" alone; proxies can be reconstructed from other fields (school attended, prior employer, even certain skill combinations correlated with regional labor markets). The more reliable control is outcome monitoring: track whether the agent's recommendations show disparate patterns across protected characteristics and investigate before a pattern becomes a discrimination claim, not after.

The control stack for a compliant hiring agent

  1. Notice at the point of interaction. A candidate or employee should know, before an AI-assisted step, that AI is involved in the process — and, for video interviews specifically, be told how the AI evaluates them and asked for consent before recording.
  2. Per-agent access scoping. A hiring agent should have access only to the candidate data relevant to its specific task — a screening agent does not need a candidate's full employment history beyond what the role requires. See least privilege for AI agents.
  3. Proxy-variable review. Audit the features an agent actually uses to rank or score candidates, not just the ones you intended it to use, for correlation with protected characteristics.
  4. Outcome monitoring for disparate impact. Track the agent's recommendation rates across demographic groups where you have the data to do so lawfully, and treat a persistent gap as a signal requiring investigation, not noise.
  5. A human decision-maker of record. Every covered employment decision needs an identifiable human who made or ratified the final call, with the agent's recommendation as an input, not the decision itself. See human-in-the-loop approvals for agents.
  6. Retention and deletion on request. For video-interview data specifically, build the deletion path the Video Interview Act requires, and treat requests as time-sensitive.
  7. A tamper-evident audit trail of the process. Be able to show, for any specific candidate, what the agent saw, what it recommended, and who made the final decision. See audit trails that hold up under regulatory scrutiny.

Why this matters beyond Illinois

Illinois has a pattern of leading on AI and biometric regulation — the Video Interview Act predates most current AI legislation, and the state's Biometric Information Privacy Act shaped how companies handle biometric data nationally well before most states had any AI-specific statute at all. The Human Rights Act amendment itself is not a preview of something coming — it has been in force since January 1, 2026. What is genuinely unsettled is the notice mechanics: IDHR withdrew its proposed rules defining when, how, and in what form notice must be given, so the statutory notice duty applies while the compliant form of notice remains undefined. The practical response is to give notice now on a conservative reading — clear, specific, given before the AI-assisted step occurs — rather than waiting for final rules that may not arrive on any predictable schedule. Treat Illinois's employment framing — notice plus non-discrimination plus a named proxy risk — as a preview of what other states are likely to adopt for employment AI specifically, distinct from the frontier-model transparency laws working through California and New York.

What good looks like

  • Candidates and employees are notified, in plain language, whenever an AI system materially shapes a step in their hiring, promotion, discipline, or termination process.
  • Video-interview candidates give informed consent before recording, and know how the AI evaluates them.
  • You have tested your hiring agent's actual feature usage for zip-code-style proxy discrimination, not just checked that protected characteristics are excluded from the input schema.
  • Every covered decision has an identifiable human decision-maker, with the agent's output logged as an input to that decision.
  • You can produce, for a specific candidate, exactly what data the agent used and what it recommended.

This is not legal advice; confirm current requirements with counsel, particularly around notice language and consent mechanics, which are the parts of these statutes most likely to be checked in practice.

Common questions

Does this apply if we use a third-party hiring platform with built-in AI features? Very likely yes — using a vendor's AI-assisted screening or interview-analysis feature makes you the employer using AI in an employment decision, and the notice and non-discrimination obligations run to you, not just to the vendor. Confirm your vendor contract addresses what data they process and what evidence they can provide if you need to demonstrate compliance.

Do we need a lawyer to review our hiring agent's prompts? A legal review is more useful applied to your process — what notice is given, when, and what a candidate can request — than to the literal prompt text. Pair legal review of the process with an engineering review of the actual features and data the agent uses to rank or screen candidates.

How is this different from a general algorithmic-discrimination law like Colorado's? Colorado's amended AI Act, covered in Colorado AI Act compliance for engineers, applies across consequential decision categories including employment, with disclosure and human-review duties. Illinois's employment-specific statutes predate and overlap with that broader framing but add employment-specific mechanics like the video-interview consent requirement that a general consequential-decision law does not include.

What is the highest-priority fix for a team that has not looked at this yet? Add explicit notice language to any candidate-facing or employee-facing step where an AI agent materially shapes the outcome, and confirm you have a named human decision-maker for each covered decision category. Those two items are the parts of Illinois's framework most directly checkable and most likely to be the subject of a complaint.

Does this only apply to employers headquartered in Illinois? No. The relevant trigger is generally where the position is based or where the affected employee or candidate is located, not where the employer's headquarters sits — consistent with how most state AI and privacy statutes scope jurisdiction. Confirm the current scope for any multi-state hiring pipeline rather than assuming Illinois-only exposure applies just because your HR team sits elsewhere.