Skip to main content
US regulatory tracker

AI hiring law in the United States, jurisdiction by jurisdiction.

There is no federal AI hiring statute. Instead there are city rules, state civil-rights amendments, general-purpose AI acts and biometric privacy laws that predate all of it. This page groups them by the obligation they create, so you can see what lands on your process without memorising statute numbers.

Book A Demo

Informational only. Not legal advice. Confirm your position with counsel.

A compliance team reviewing jurisdiction-specific AI hiring obligations
One operational view across changing state and city rules.
How to read this page

Six obligations. Every instrument is some combination of them.

Statute names change from state to state. The duties underneath barely do. Once you can answer these six questions about your process, adding a jurisdiction is a scoping exercise rather than a rebuild.

Notice

Tell the candidate that software is evaluating them, usually before the round rather than after. Some rules add what the tool assesses and what data it collects, and some require the notice to be publicly posted rather than only sent to the applicant.

Consent

Get an affirmative yes before the analysis runs. This is the core of the Illinois video interview regime and of every biometric privacy statute, and it usually implies you have an alternative route for anyone who declines.

Testing or audit

Show that the tool was checked for discriminatory impact. The demands vary sharply, from an independent annual bias audit with a published summary, to an impact assessment held on file, to a general duty of reasonable care.

Records

Keep the applications, the criteria the system used, the inputs and the outputs, for a defined period. This is the most common reason an employer cannot defend a decision that was fine.

Human review

A person who can review, override and stop the system, and in some frameworks an explicit route for the candidate to ask for human reconsideration. Rarely a full exemption on its own, but it lowers exposure almost everywhere.

Deletion

Destroy the recording or the identifier on request, within a stated window, including copies held elsewhere. Deletion that stops at the interface and leaves backups intact does not satisfy this.

Where we go deep

Three jurisdictions have a full guide behind them.

These are the ones an employer running AI-assisted hiring in the United States hits first, and each has a dedicated page with the scoping questionnaire, the split of responsibilities between vendor and employer, and what you have to produce.

New York City

Enforced since July 2023

Local Law 144

Covers automated employment decision tools used to screen candidates for roles based in the city. The distinguishing feature is that the testing obligation is external and public: an independent auditor, annually, with a summary you have to publish. Notice to candidates has to come before the tool is used, and it has to say what data is collected.

Independent annual bias audit, published summary, advance candidate notice.

Read the guide

Illinois

AIVIA in force since 2020

AI Video Interview Act, plus the Human Rights Act amendment

The video interview act is consent-first: notice and an explanation of what the AI evaluates, an affirmative yes before the analysis runs, strict limits on who may see the recording, and destruction within 30 days of a request. A separate amendment to the state's Human Rights Act extends discrimination liability to AI used across employment decisions and adds its own notice duty.

Consent before analysis, sharing limits, deletion within 30 days of request.

Read the guide

California

In force

FEHA rules on automated decision systems

California did not create a separate AI regime for hiring. It confirmed the existing anti-discrimination law already reaches automated decision systems, which puts the employer on the hook for outcomes rather than intentions. Vendor assurance is generally not a defence you can lean on, pre-deployment testing matters, and record retention is longer than most teams assume.

Pre-use testing, long record retention, employer keeps the liability.

Read the guide
The reference

What is on the books, and what it asks for.

Described at the level of the obligation rather than the clause. Where an instrument has a contested scope or a commencement date that has already moved, the row says so instead of asserting a date we would have to keep correcting.

JurisdictionInstrumentWhat it coversObligationsStatus
New York CityNYC guideLocal Law 144Automated employment decision tools used to screen candidates for hiring or promotion, where the role is based in New York City.NoticeTesting or auditRecordsIn force. Enforced since July 2023.
IllinoisIllinois guideAI Video Interview Act (820 ILCS 42)AI analysis of video interviews submitted by applicants for positions based in Illinois.NoticeConsentDeletionIn force since 2020.
IllinoisIllinois guideHuman Rights Act amendment on AIAI used across recruitment, hiring and other employment decisions, treated through the state's existing discrimination framework rather than as a standalone AI statute.NoticeTesting or auditHuman reviewIn force. Rulemaking has followed commencement, so confirm the current position.
CaliforniaCalifornia guideFEHA regulations on automated decision systemsAny automated decision system used to screen, rank or evaluate applicants, brought inside the state's existing anti-discrimination law rather than regulated separately.NoticeTesting or auditRecordsHuman reviewIn force.
CaliforniaCalifornia guideState privacy rules on automated decision-making technologyAutomated decision-making technology used for significant decisions, including employment, under the state privacy regime rather than employment law.NoticeRecordsHuman reviewAdopted with phased compliance dates that run beyond the employment rules. Confirm which phase applies to you.
ColoradoColorado AI Act (SB 24-205)High-risk AI systems that make or substantially factor into consequential decisions, employment expressly among them. Splits duties between developers and deployers.NoticeTesting or auditRecordsHuman reviewEnacted. The commencement date has been moved once already, so treat the effective date as a live question and confirm it with counsel.
MarylandH.B. 1202Use of facial recognition services during a pre-employment interview. Narrow, and squarely aimed at face analysis rather than AI generally.ConsentIn force since 2020.
TexasGeneral-purpose state AI statuteDevelopment and deployment of AI systems across sectors, with an intent-based prohibition on unlawful discrimination rather than an employment-specific compliance regime.NoticeRecordsIn force. Scope is general rather than employment-specific, so the employment analysis runs through existing discrimination law.
New JerseyAttorney General guidance on algorithmic discriminationExisting state anti-discrimination law applied to algorithmic decision tools, including in hiring. Guidance rather than a new filing or audit duty.Testing or auditHuman reviewGuidance in effect. No separate registration or audit filing created.
Federal (United States)Title VII, ADA and ADEANo AI-specific federal hiring statute exists. Existing anti-discrimination law applies to a selection procedure regardless of whether a human or a model ran it.Testing or auditRecordsHuman reviewIn force. Federal agency guidance on AI has changed with administrations. The underlying statutes have not.

Scroll the table sideways to see every column.

This table is not exhaustive and it is not legal advice. It is provided for informational purposes only and does not constitute legal advice. Obligations vary based on how tools are configured and used, instruments are amended, and effective dates move. Employers are responsible for their own compliance decisions. Confirm your position with qualified counsel before you rely on anything here, and tell us if you spot something out of date.

The layer underneath

Biometric privacy law reaches recorded interviews.

This is the part teams miss. A recorded interview can contain a voiceprint and face geometry, and several states regulate those as biometric identifiers under privacy law. These rules predate the AI debate by more than a decade and they apply whether or not a model is involved.

Illinois

Biometric Information Privacy Act (740 ILCS 14)

Written notice and written release before a private entity collects a biometric identifier, plus a published retention and destruction schedule. Voiceprints and face geometry are named identifiers, which is why it reaches recorded interviews. It carries a private right of action, which is what makes it the most litigated statute in this group.

Texas

Capture or Use of Biometric Identifier statute

Informed consent before capturing a biometric identifier for a commercial purpose, with retention limits. Enforced by the attorney general rather than by private suit, which changes the risk profile but not the obligation.

Washington

State biometric identifiers statute

Notice and consent before enrolling a biometric identifier in a database for a commercial purpose. Narrower in scope than the Illinois act and without an equivalent private right of action.

Practical read: if you record interviews and hire anywhere on this list, your notice and consent step has to satisfy the biometric statute as well as the AI rule, and your retention schedule has to be written down. Settle both before the first candidate, not during discovery.

Where the law meets the config

Which of your rounds is automated is the whole question.

Almost every instrument on this page is scoped by what the software does, not by what the vendor is called. That means the compliance answer is decided in the round builder, by the person setting up the pipeline, weeks before anyone from legal is in the room.

The round builder, shown with sample data. Step 1 of 3 on the left, and the five modes a stage can be: Resume Screening, AI Interview, Human-Led Interview, AI Prompt Assessment and Coding Test. Each card states what that round evaluates. The prompt at the top reads pick how this stage should evaluate candidates, which is the sentence your counsel will want written down.

The scoping question is per round, not per platform.

Resume Screening evaluates without an interview. AI Interview runs a live voice round. Human-Led Interview is a recruiter call with a notetaker. Prompt assessment and coding test are scored exercises. Those are five different answers to the question a regulator asks, and a single job can mix them.

Write down what each round is allowed to decide.

Whether a score advances a candidate on its own or whether a person decides is what most of these instruments turn on. Settle it while you are configuring the round, when it costs a sentence, rather than after a complaint, when it costs a defence.

The rubric is set here and applies to everyone.

The builder walks Basic Details, Round Configuration and Questions, so every candidate in a round faces the same structure and is scored on the same basis. That consistency is the thing a disparate-impact analysis needs in order to say anything useful.

A workable posture

One process, run at the highest bar.

You cannot maintain fifty variants of a hiring process. These four moves satisfy the common core of nearly everything on this page. Only the local obligations, such as the New York City audit and its published summary, are handled separately.

  1. 01

    Pick the strictest bar you are exposed to.

    Branching a hiring process by state is expensive to build and worse to maintain, and it fails the moment a remote requisition opens. Run one process that clears the highest bar among the jurisdictions you hire in, then add the local extras on top.

  2. 02

    Give notice before the round, every time.

    Notice is the cheapest obligation on the list and it appears in almost every instrument. Candidates review a plain-language AI notice and consent before a video round begins. Doing it everywhere costs nothing and removes a whole category of argument.

  3. 03

    Keep a person between the score and the outcome.

    Scores are advisory until someone acts on them. Malpractice and integrity signals are surfaced to a human reviewer alongside the transcript and the code. There is never an automated pass or fail, and we do not auto-reject anyone.

  4. 04

    Retain the record for the longest period in scope.

    Different states expect different retention windows. Pick the longest one that touches you and apply it everywhere. Audit logs sit on every action, reports keep the transcript and the recording, and candidate data exports to CSV whenever an auditor wants the underlying record.

The security and data-handling side of this posture is documented in our technical and organisational measures and our privacy policy. The vocabulary these instruments use is defined in the glossary, and the wider jurisdiction map lives on the compliance hub.

FAQ

What people ask about the patchwork.

Seven questions that come up whenever a team hires across more than one state.

Still unsure how this lands on your setup? Talk to compliance support and we will walk your configuration with you.

Hire across states without rebuilding the process.

Structured rounds, notice and consent inside the interview flow, human review before anyone moves, and a record you can export. Start on the free trial and see what the reportcard gives your legal team.

30-day free trial with 75 Hire Credits. Card required, nothing charged for 30 days.

Prefer to start with the mechanics? Read how scoring works or see the platform.