New York City's Local Law 144 generally requires a bias audit and notice to applicants before an employer uses an automated employment decision tool, and enforcement began July 5, 2023. Illinois amended its Human Rights Act, effective January 1, 2026, to require notice when employers use AI in employment decisions. If your team screens applicants with an AI tool and hires in either place, you need to know what each rule asks and when it started applying.
Sharp takeaways
- New York City's rule attaches before an automated employment decision tool is used, and enforcement began July 5, 2023.
- Illinois's notice requirement has applied since January 1, 2026, and it is tied to discriminatory effects of AI in employment decisions.
- Both rules put notice to applicants at the center, so a clear disclosure is worth building once and reusing.
- EEOC regulations (29 CFR 1602.14) generally require keeping hiring records for one year, counted from the later of the record date or the personnel action.
New York City's Local Law 144 and when enforcement began
New York City's Local Law 144 is the first of the two rules, and its core structure is simple to state. According to GoodHire's guide to applicant tracking systems, the law generally requires two things before an employer uses an automated employment decision tool: a bias audit and notice to applicants. Both duties sit upstream of the decision. They are not a cleanup task to handle after shortlists are built, which is why teams should map them onto the point in their workflow where software first touches an applicant.
The word "generally" in that summary deserves attention. It signals that the rule has conditions, and a short overview cannot capture every scenario. Read the summary as the core obligation, then work out for your own process which tools fall within it. For a small staffing team, the practical question is usually whether the software you already use to sort, score, or rank applicants is doing work that would otherwise be done by hand.
The city began enforcing the law on July 5, 2023. That date gives the rule a fixed point in time. If your team adopted a screening tool before that date, the enforcement date is still the reference for when the rule became live in New York City, so it is the date to compare against your own adoption records.
Timing also shapes documentation. Because the audit and notice duties are tied to use, a team should be able to show which tool was in service, when it was adopted, and what applicants were told. Keep a dated record of each screening tool and of the notices you sent. That record costs little to maintain, and it gives you a direct answer when an applicant or auditor asks how applications were screened.
Illinois notice rule effective January 1, 2026
Illinois took a different route. Illinois took a different route: the state amended its Human Rights Act. GoodHire's guide reports that the amendment took effect January 1, 2026, that it addresses discriminatory effects of AI in employment decisions, and that it requires notice when employers use AI. Placing the rule inside a civil rights statute means the central concern is the outcome of AI-assisted decisions, not only the disclosure. A team using a screening tool in Illinois has to think about whether the tool produces unequal results as well as whether applicants were told.
The four facts that matter are set out below.
| Element | What the source says |
|---|---|
| Law amended | Illinois Human Rights Act |
| Effective date | January 1, 2026 |
| Core concern | Discriminatory effects of AI in employment decisions |
| Required step | Notice when an employer uses AI in employment decisions |
The effective date is the one to act on. The rule has applied since January 1, 2026, so it is not a future obligation. An AI tool your team uses to make or shape employment decisions about applicants in Illinois falls under a rule that is already in force, and the notice element is the part most visible to applicants.
Both laws put notice at the center of compliance, so an applicant-facing disclosure is a shared element worth building once. The difference lies in what sits behind the notice. New York City's rule pairs notice with a bias audit. Illinois's rule frames notice around discriminatory effects. A single disclosure can serve both places, but the underlying evidence your team gathers about the tool should be strong enough to answer questions about results.
How long to keep hiring records
Notices and audits are only useful if the underlying records still exist. GoodHire's guide cites EEOC regulations at 29 CFR 1602.14, which say employers covered by federal anti-discrimination laws generally must keep application forms and other hiring records for one year from when the record was made or the personnel action was taken, whichever is later.
The key facts from that rule are collected in the table below.
| Item | Rule from the source |
|---|---|
| Citation | 29 CFR 1602.14 (EEOC regulations) |
| Who it generally covers | Employers covered by federal anti-discrimination laws |
| Records covered | Application forms and other hiring records |
| Retention period | One year |
| Start of the clock | Later of when the record was made or when the personnel action was taken |
The "whichever is later" clause changes the arithmetic. A record created when an applicant applies may need to be kept longer if a hiring decision follows afterward, because the clock starts from the later event. A team that deletes screening data a few months after an application arrives may be discarding records that are still within the period.
For AI screening, the practical step is to treat a tool's scores, rankings, and screening outputs as part of the file. Treating these outputs as part of the file is a conservative practice, since the source does not say whether the regulation covers them. Set a retention schedule that counts from the later event, not the application date, and apply it to the tool's outputs as well as to the resumes that went into it.
What the 2026 Stanford study found about AI screening
A 2026 Stanford study examined one vendor's AI screening tool across about 4 million applications. GoodHire's guide reports that the study found racial disparities in the tool's recommendations when postings were analyzed one at a time. That finding is specific in two ways. It concerns one product, and it depends on how the analysis was grouped. A pattern that appears when each posting is examined separately is a reason to ask how your own data is cut.
The scale is what makes the study difficult to wave away. A 2026 Stanford study of one vendor's AI screening tool covered about 4 million applications. Still, the study is about one vendor's tool. Your team's software may be built differently, trained on different data, or configured differently for each role you recruit for. The study is best read as a warning about what to test for, not as a verdict on every product in the market.
For a recruiter, the lesson concerns verification more than reputation. Ask the vendor what testing it has run, how its results were grouped, and whether it reviewed postings individually or only in aggregate. New York City requires a bias audit before an automated employment decision tool is used, so the audit and the vendor's own analysis should be read together.
Document the answers. If the vendor cannot explain how recommendations were checked, that gap belongs in your records alongside the tool's outputs. A team that can show it asked the right questions is in a stronger position than one that simply accepted a ranked list.
Recruiter stress over AI-polished resumes
Lever's recruiting statistics for 2027 report that AI-polished resumes are eroding trust in the resume itself, and that 33% of recruiters name them a top source of stress. The figure points to a broader problem. Lever reports that AI-polished resumes are eroding trust in the resume and that 33% of recruiters name them a top source of stress. When many applicants' resumes read as clean and well organized, polish stops distinguishing one applicant from another.
That uncertainty travels into screening. A tidy document no longer reliably indicates the skills or experience behind it. A screening tool that reads resumes inherits the same doubt, and a tool that ranks polished documents may reward presentation over substance.
The response is to gather evidence inside your own process rather than relying on the resume alone. Structured questions asked of every applicant, and short work samples where they fit the role, give you comparable information. Those steps also produce the kind of consistent records that the retention rule and the notice duties depend on.
Why AI adoption is stalling
Lever's same report describes a gap between using AI and getting results from it. It says 68% of teams use AI, but only 18% have shifted more than half their time to high-value work, down from 26% in 2025. Those figures answer a different question than adoption rates do. Adoption asks whether a team uses the tool. Impact asks what the team does with the hours the tool saves.
The three figures from the report are set out in the table below.
| Measure | Figure from Lever |
|---|---|
| Teams that use AI | 68% |
| Teams that shifted more than half their time to high-value work | 18%, down from 26% in 2025 |
| Recruiters naming AI-polished resumes a top source of stress | 33% |
Read together, the figures show where the gap sits. Most teams have the tool, but a much smaller share has moved its time toward judgment-heavy work such as interviews, applicant conversations, and hiring-manager alignment. A screening tool that only speeds up sorting has not changed how the team spends its week. The decline in the share shifting over half their time, from 26% in 2025 to 18%, suggests the gap has not closed.
For a staffing team, this matters because the compliance duties follow the tool, not the team's intentions. A team that adopts a screening tool without measuring what it changes will struggle to explain why the tool is worth the notice, audit, and record-keeping work it brings. Track where the hours go after adoption, and keep that tracking alongside the records you already maintain.
Applicant fraud in 2026
Lever reports that the share of organizations confirming applicant fraud rose from 23% in 2025 to 27% in 2026. Lever's recruiting statistics for 2027 report that 27% of organizations confirmed applicant fraud in 2026, up from 23% in 2025. That is a four-percentage-point rise in a single year, and it has arrived at the same time that AI tools are becoming part of how applications get read.
The connection matters for screening. A tool that sorts or ranks applicants works from the information applicants supply. If that information is inaccurate, the ranking can be confidently wrong. Automated screening does not verify a claim just because it arrives in a clean format. Lever's report also describes AI-polished resumes eroding trust in the document itself, which makes this risk harder to spot.
For a small staffing team, the response is verification built into the process, not suspicion aimed at individual applicants. Decide in advance which claims you will check, such as identity, licensing, and employment history for roles where those details carry weight. Apply the same checks to every applicant at the same stage so the process stays consistent. Then record what was checked and when. That record keeps the file useful if a hiring decision is later questioned.
Keep the wording careful, too. The source reports a rise in organizations that confirmed fraud. It does not measure how common fraud is among all applicants. A team should not assume that every applicant is dishonest. The stronger approach is to make verification routine, so that it protects honest applicants as much as it catches false claims.
Keeping interviews consistent and fair
Interviews are where a screening tool's output gets tested against a real person. Workable's guide to interview questions describes prepared questions as rich, researched, and ready to go, and says that when questions are prepared this way, they keep momentum and keep the team's approach consistent, fast, and fair. That is a useful standard for any recruiter who asks questions by hand, and it applies equally well to a team that also uses software earlier in the process.
Consistency is the core idea. If one applicant answers a set of prepared questions and another has a loose conversation that wanders, the two sets of answers cannot be compared fairly. Fairness depends on the same questions being asked the same way. Speed depends on questions being ready before the interview starts rather than invented in the room.
This matters for AI screening in two ways. First, the notice duties in New York City and Illinois concern notice to applicants about the use of these tools. A structured, documented interview stage makes it easier to show that the human part of hiring was deliberate rather than a rubber stamp on a ranked list. Second, the federal retention rule covers hiring records generally. Interview notes, scoring sheets, and the question set used for each role belong in the file, and they are easier to keep when they are standardized.
A practical starting point is a question bank for each role family, reviewed before postings open. Write questions tied to the job's actual duties, prepare a short scoring guide for each one, and use the same set for every applicant at the same stage. Revise the bank when the role changes, and note which version was used for each interview.
Side-by-side: New York City and Illinois
The two rules share a goal of making AI use in hiring visible to applicants, but they differ in form, timing, and emphasis. GoodHire's guide describes New York City's Local Law 144 as generally requiring a bias audit and notice to applicants, with enforcement beginning July 5, 2023. Illinois instead amended its Human Rights Act, effective January 1, 2026, to address discriminatory effects of AI in employment decisions and to require notice when employers use it. The comparison is set out in the table that follows.
| Feature | New York City (Local Law 144) | Illinois (Human Rights Act amendment) |
|---|---|---|
| Type of rule | Local law | Amendment to the Human Rights Act |
| Date it applies from | Enforcement began July 5, 2023 | Effective January 1, 2026 |
| Core requirements | Generally a bias audit and notice to applicants | Notice when employers use AI in employment decisions |
| Focus named in the source | Automated employment decision tools | Discriminatory effects of AI in employment decisions |
The biggest difference is emphasis. New York City pairs notice with an audit requirement, so evidence about the tool becomes a formal part of compliance. Illinois frames its rule around outcomes, which means a team hiring there should be able to explain what the tool does for different groups of applicants, not only that applicants were told. Both laws put notice at the center, so one disclosure can be written and then adjusted for each location.
The Illinois rule is also more recent. Teams should not assume that practices copied from the New York City approach automatically satisfy Illinois. Check each requirement against its own source rather than treating the two as interchangeable.
Which one should you pick?
For a team that hires in both New York City and Illinois, the real decision is not which law to follow. Both apply to the tools you use in those places. The useful choice is how to organize your process so both are covered without running two separate systems.
Compare two options. The first is a separate process for each city, with its own notice wording, its own checklist, and its own records. That approach creates more paperwork and more chances for an applicant in one place to receive the wrong notice. The second is one shared process: a single notice stating that an AI tool is part of screening and describing what it does, placed in the application flow before any tool reads a resume, plus a New York City bias audit step for roles covered by that rule. For a small staffing team, the second option is the stronger pick, because it leaves fewer moving parts and one set of records to maintain.
Pair the shared notice with the consistent interview process described in Workable's guidance. That keeps the human stage documented, so a ranked list is never the only evidence behind a decision. Then have counsel review the notice wording and confirm how each law applies to the specific tools you use.
Figures at a glance
Two of Lever's 2027 figures moved in directions recruiters should watch, and one figure has no earlier comparison. The table below gives the earlier and later values where the source provides them.
| Measure | Earlier figure | Later figure |
|---|---|---|
| Organizations that confirmed applicant fraud | 23% (2025) | 27% (2026) |
| Teams that shifted more than half their time to high-value work | 26% (2025) | 18% (2026) |
| Teams that use AI | Not given in the source | 68% |
The first row rose four percentage points. The second fell eight percentage points, from 26% to 18%. Lever's report frames the second figure as a sign that AI adoption is stalling: most teams have the tool, but fewer have turned it into time spent on higher-value work. The 68% adoption figure has no earlier comparison in the source, so it should not be read as a trend.
Lever also reports that 33% of recruiters name AI-polished resumes a top source of stress. That figure carries no year in the source, so it stands on its own rather than as a comparison.
For a staffing team, these numbers argue for measuring rather than assuming. Track how much time screening actually saves, how often verification catches a problem, and whether interview scoring stays consistent across applicants. Those measures give you a record you can show, and they tie back to the notice and retention duties covered earlier in this post.
Related guides: AI screening agents: 4 vendor figures to check · State wage law coverage: 4 laws before an offer · Screening criteria: 3 groups to set before hiring.
Reader questions
Does a bias audit replace the notice to applicants? No. According to GoodHire's guide, New York City's rule generally requires both a bias audit and notice to applicants before an automated employment decision tool is used.
When did New York City begin enforcing its rule? The city began enforcing Local Law 144 on July 5, 2023.
Does Illinois's rule apply now? Yes. The amendment to the Human Rights Act took effect January 1, 2026, so it is already in force rather than a future obligation.
Does high AI adoption mean recruiters have more time for judgment work? Not necessarily. Lever reports that 68% of teams use AI, but only 18% have shifted more than half their time to high-value work, down from 26% in 2025.
Is applicant fraud becoming more common? Among organizations, the share that confirmed it rose from 23% in 2025 to 27% in 2026, according to Lever. The source does not measure how many individual applicants commit fraud.
What makes interview questions support a fair process? Workable describes prepared questions that are rich, researched, and ready to go. It says those questions keep momentum and the team's approach consistent, fast, and fair.
Are AI-polished resumes a problem for recruiters? Lever reports that they are eroding trust in the resume itself, and that 33% of recruiters name them a top source of stress. The source does not say how many resumes are affected.
Not legal or tax counsel.
Using AI software to screen job applicants? AI Hiring Paperwork Pack is a workbook filled from your answers that shows which rules reach each tool and the wording applicants should get.