What Your AI Notice Must Say, and When Each Jurisdiction Wants It

What does New York City require, and how far ahead?

Ten business days, in writing, before the tool runs on that candidate. Local Law 144, in force since January 1, 2023 and enforced since July 5, 2023, bars an employer from using an automated employment decision tool on a candidate in the city unless a bias audit was completed within the prior year, a summary of the results is posted publicly, and the candidate received notice at least ten business days ahead ¹. Business days, not calendar days.

The notice itself has required contents. It has to say that an automated tool will be used, name the job qualifications and characteristics the tool assesses, and include instructions for requesting an alternative selection process or a reasonable accommodation, where one is available; the rules state expressly that nothing requires an employer to provide an alternative selection process ⁶. A standing notice on the careers page satisfies the rule, and it starts the clock for everyone who reads it.

Two limits bound the duty. It attaches only where the tool substantially assists or replaces discretionary decision-making, and an employer that concludes its tool merely informs a human reviewer takes itself out of scope without asking anyone ¹. And enforcement has been thin: a New York State Comptroller audit of the first two years found DCWP had received two complaints in the whole period, with nine of the auditors' twelve test calls to 311 never reaching the agency at all ⁴.

Do not read too much into the thin enforcement. A broken complaint line says more about 311 than about employers, and it is a poor reason to skip a duty that costs a page of text. It does mean the pressure to comply is more likely to arrive through a discrimination claim than through a city inspector.

Why is Illinois the widest trigger and the thinnest instruction?

Because the statute defines artificial intelligence broadly and then says almost nothing about the notice. Since January 1, 2026 the Illinois Human Rights Act has made it a civil rights violation to use AI that has the effect of discriminating in recruitment, hiring, promotion or discharge, to use zip codes as a proxy for a protected class, or to fail to notify an employee that AI is being used for those purposes ².

The definition is the part that sweeps. Artificial intelligence there means a machine-based system that infers from its input how to generate outputs such as predictions, content, recommendations or decisions that can influence physical or virtual environments, and it expressly includes generative AI ². That reaches a sourcing tool, a resume ranker, a scheduling assistant that decides who gets a slot, and a manager comparing candidates in a chat window. If you are trying to work out whether your existing stack is inside it, start with whether your ATS counts as AI at all.

The notice instruction never arrives. The same subsection directs the Department of Human Rights to adopt rules covering the circumstances and conditions that require notice, the time period for providing it, and the means ². Until those rules exist, the act itself fixes neither the timing nor the contents. The prohibition and the effective date are in the statute regardless.

Illinois already has one AI notice duty that fixes both. The Artificial Intelligence Video Interview Act has required an employer using AI to analyse a recorded video interview for an Illinois position to notify the applicant beforehand, explain how the AI works and what general types of characteristics it evaluates, and obtain the applicant's consent, since 2020 ⁵. It bars evaluating an applicant who has not consented, and on request the video has to be deleted within 30 days. It reaches that one technology and nothing else, so it says nothing about a resume screener or a take-home.

One drafting detail belongs with counsel. The 2026 act's notice clause runs to an employee, and Article 2 of the Act defines an employee as a person performing services for remuneration, an apprentice, or an applicant for an apprenticeship ². An outside job applicant is not plainly inside that list, even though the discrimination clause above it plainly covers recruitment and hiring. Notify applicants anyway: the cost of the wider reading is a sentence and the cost of the narrower one is an argument.

What does California want instead of a notice?

Records. California's amended employment regulations brought automated-decision systems inside the discrimination rules on October 1, 2025, defining such a system as a computational process that makes or facilitates a decision about an employment benefit, and naming resume screening and the analysis of facial expression, word choice or voice in online interviews as examples ³⁷. Nothing in them requires telling a candidate anything.

What they require instead is that you can reconstruct the decision later. The same amendments extended the employment-records retention period from two years to four and said expressly that automated-decision system data is included ³. They also made anti-bias testing legally relevant without mandating it: in a claim that such a system discriminated, evidence or the absence of evidence of testing is weighable, along with its quality, recency, scope, results, and what the employer did about the results ³. FEHA reaches employers with five or more employees.

That last clause has a sharp edge. An employer that tests, finds a disparity and does nothing has manufactured the exhibit against itself. Testing is still the right call, but budget for acting on what it says.

One more clause changes who is liable. An agent acting for an employer, including a vendor running the system, is itself an employer under the Act ³. Buying the tool does not move the duty to the seller. It adds a second defendant.

Write one notice to the widest trigger

One document, delivered before the tool runs, given to every candidate rather than only the ones in a covered state. Sorting candidates by jurisdiction at the moment of notice costs more than the notice does, and it breaks the first time someone relocates mid-process. Draft to the widest trigger and the strictest timing you are subject to, then let counsel narrow it.

Eight things belong in it, and all of them are things you already know:

• What the tool is, in the words a candidate would use, with the vendor's product name second.
• What stage it is used at, and what it does at that stage.
• What it assesses: the job qualifications and characteristics, stated concretely.
• What categories of candidate data it takes as input.
• Whether a human reviews the output before a decision, and at what point.
• A named contact, with a working address, for questions about the process.
• How to request an alternative selection process or a disability accommodation.
• When the notice was given, recorded per candidate.

Timing: ten business days before first use for anyone who might be a New York City candidate, which in a remote-hiring pipeline is most of them. Consent as well, if AI will analyse a recorded video interview for an Illinois position, captured as an affirmative yes before anyone evaluates it. Language: plain, and in the languages your applicant pool actually reads. Format: the same text in the posting, in the application flow, and in the file, so the version you produce later is the version they saw.

The notice is also the cheapest half of a harder problem. When someone is rejected and asks what happened, the notice covers what was used, and what it concluded is a separate conversation with its own duties. That one is worked through in what a rejected candidate is owed when AI screened them out, and the sign-off, ownership and review questions around all of it sit in what actually belongs in an AI hiring policy.

Last thing. If a manager has already been pasting resumes into a general chatbot on a live requisition, your notice is already late for that req. Say so internally before the notice goes out.

Read the evidence >