Are AI Job Interviews Legal? US State Rules for Employers in 2026

AI job interviews are legal in the United States, but four jurisdictions impose specific duties on the employer, and it is the employer, not the vendor, who is liable. New York City requires an independent bias audit and 10 business days of candidate notice. Illinois requires notice, explanation and consent for AI-analyzed video interviews. Maryland requires a signed waiver before facial recognition is used. Colorado has a broad AI law arriving January 1, 2027. Everywhere else, Title VII and the ADA still apply in full. This guide sets out what each rule actually requires, what it costs to get wrong, and the practices that keep an AI screening round defensible in any state.

The confusion in this area comes from two directions. Vendors tend to describe compliance as a feature they provide, and roundup articles tend to repeat effective dates that have since moved. Both are misleading. No vendor can discharge your obligations for you, because the laws attach to the employer using the tool, and at least one widely cited state deadline changed twice in the last two years. What follows is current as of July 2026.

Is it legal to use AI to interview job candidates?

Yes. There is no federal law banning AI interviews or AI candidate scoring, and no state has prohibited them either. What exists instead is a layer of disclosure, consent and audit requirements in specific jurisdictions, sitting on top of the anti-discrimination law that already governed hiring. If your AI screening process produces a disparate impact on a protected group, that is unlawful under Title VII regardless of which state you are in and regardless of whether any AI-specific statute applies.

The EEOC has been clear that using a vendor's tool does not transfer responsibility. An employer that screens candidates with software it did not build is still the party that made the employment decision, and still the party that answers for the outcome. Practically, this means the questions to ask a vendor are evidentiary: can they show you the bias audit, will they hand over the scoring criteria, and can you export the records you would need if someone challenged a rejection.

Which states regulate AI interviews?

New York City: the strictest regime

Local Law 144 governs what it calls an automated employment decision tool, or AEDT. The definition is broader than most employers assume: any computational process derived from machine learning, statistical modeling, data analytics or AI that issues a simplified output such as a score, classification or recommendation, and that substantially assists or replaces discretionary decision making. A weighted spreadsheet can qualify. Whether the tool uses AI in the marketing sense is not the test.

Three duties follow. First, an independent bias audit conducted within the previous 12 months, testing sex, race and ethnicity categories and their intersections. Independent means not conducted by you and not by the vendor. Second, a publicly published summary of that audit. Third, notice to candidates at least 10 business days before the tool is used. We cover the definition and the audit mechanics in more detail in our guides to what counts as an AEDT and how a Local Law 144 bias audit works.

Penalties start at 500 dollars for a first violation, with up to 500 dollars more for each additional violation on the same day, and 500 to 1,500 dollars for each subsequent violation. The detail that turns this from a nuisance into a real exposure is that each day of unlawful use counts as a separate violation, and a failure to give notice counts as its own separate violation on top of that. A tool running unaudited across a hiring quarter accumulates penalties daily.

Why NYC enforcement is about to tighten

Enforcement has been light so far, and that is changing. On December 2, 2025, the New York State Comptroller published an audit of the Department of Consumer and Worker Protection covering July 2023 to June 2025. Over that period DCWP had reviewed 32 companies and identified one issue. The Comptroller examined the same set of companies and found at least 17 potential violations. The audit also found that roughly 75 percent of 311 test calls about AEDT complaints were misrouted, meaning complaints were not reliably reaching the enforcing agency. It made 13 recommendations, of which DCWP fully adopted 10.

The reasonable read for an employer is that the low historical enforcement rate reflected a detection gap rather than widespread compliance, and that the gap is being closed. Building the audit trail now is considerably cheaper than reconstructing one under a complaint.

Illinois, Maryland and Colorado

Illinois has had rules since 2020. The AI Video Interview Act applies when an employer uses AI to analyze video interviews for Illinois positions, and requires notice before the interview, an explanation of how the AI works and what characteristics it evaluates, and consent from the candidate. Candidates can also require destruction of their video within 30 days of request. Separately, HB 3773 amended the Illinois Human Rights Act effective January 1, 2026, making it a civil rights violation to use AI that discriminates on protected characteristics, and specifically naming zip code as an impermissible proxy.

Maryland's HB 1202 is narrower and easier to comply with: get a signed waiver before using facial recognition that creates a facial template during an interview. If your screening tool does not analyze faces, the requirement does not bite.

Colorado is the one most commonly reported wrong. Many 2026 roundups still say the Colorado AI Act takes effect June 30, 2026. It does not. SB 24-205 was delayed to that date by SB 25B-004, signed August 28, 2025, and was then replaced by SB 26-189, signed May 14, 2026, with an effective date of January 1, 2027. If your compliance calendar has a Colorado deadline in mid-2026 on it, it came from a stale source.

Do I have to tell candidates they are being interviewed by AI?

In New York City, Illinois and Maryland, yes, in the specific forms those laws describe. Everywhere else there is no statutory disclosure duty for AI screening. Disclose anyway. The cost is a sentence in the invitation email, and the downside of not disclosing is a candidate who feels deceived, which is the fastest route to a public complaint and to the kind of reputational story that outlasts any efficiency gain.

Practical disclosure has four parts: tell the candidate before they start that AI will conduct or analyze the interview, explain in plain language what it evaluates, confirm that a human reviews the results and makes the decision, and offer a route to request an alternative format. That last one also does double duty as your ADA accommodation path, which is a genuine legal requirement in every state. A candidate with a speech difference, a hearing impairment or a condition affecting response timing may need an alternative, and refusing one is a far more likely source of liability than the AI itself.

What makes an AI interview process defensible?

The same practices satisfy nearly every rule above, which is convenient, because building to the strictest standard once is cheaper than tracking five regimes. Six things matter.

Ask every candidate for a role the same structured questions, scored against a documented rubric written before anyone applied. Score what candidates say, not how they look or sound, and avoid any tool inferring traits from facial expression or vocal tone, which is both the weakest science in this field and the fastest way into Maryland and Illinois trouble. Keep a human decision-maker who reviews the shortlist and can override it. Run an independent bias audit annually and publish the summary. Disclose and get consent before the interview. And retain the transcripts, scores and rubric, because the defense to a discrimination claim is documentation, and reconstructing it later is not possible.

If your organization is tracking obligations across several states at once, it is usually worth putting the recurring duties on a system that watches deadlines for you rather than a spreadsheet somebody updates when they remember, because the audit-refresh dates and notice periods are exactly the kind of obligation that quietly lapses.

Does an AI interview tool make my company compliant?

No, and this is the most expensive misunderstanding in the category. Compliance obligations attach to the employer. A vendor can make compliance achievable by conducting bias audits, publishing results, providing disclosure templates and keeping exportable records, and a vendor that does none of those makes compliance nearly impossible. But the audit that Local Law 144 requires must be published by you, the notice must come from you, and the discrimination claim will name you.

What to require of any vendor, in writing: a current independent bias audit you may review, plain documentation of what the tool measures and how it scores, confirmation that it does not analyze facial expression or vocal tone, exportable transcripts and scores, and a contractual commitment to notify you when the model materially changes. A vendor that will not answer those questions is telling you something.

InterviewAgent.ai conducts structured first-round interviews where every candidate for a role gets the same questions, scores are tied to a rubric you define, and the agent scores only the content of answers. It never rejects a candidate on its own; a recruiter reviews the ranked shortlist and decides. You can see how the round is built on our automated screening interview page, or read how the scoring works on candidate interview scoring.

Frequently asked questions

Can an AI reject a candidate automatically?

Technically yes, legally unwise. Fully automated rejection is the fact pattern regulators and plaintiffs are most interested in, because it removes the human judgment that anti-discrimination law assumes exists. Keeping a person in the loop who reviews the shortlist and can override the ranking materially reduces risk, and costs very little, since reviewing a scored shortlist takes minutes.

Does Local Law 144 apply if my company is not in New York City?

It applies based on where the job is, not where you are headquartered. If you are hiring for a role located in New York City, or a remote role that a New York City resident could hold, the safer assumption is that it applies. Many multi-state employers simply build to the New York City standard everywhere, because running two processes is more expensive than running the strict one.

How much does a bias audit cost?

It varies with the tool's complexity and the volume of candidate data being tested, and independent auditors price per engagement rather than from a list. The larger practical constraint is not cost but data: an audit needs enough candidates across the tested categories to produce meaningful impact ratios, so a low-volume employer may need the vendor's aggregate audit rather than one on its own hiring data alone.

Are AI interviews biased?

They can be, and so can human interviews. The relevant question is which is measured. An unstructured human screen produces no record of what was asked, no consistent scoring, and no audit trail, so its bias is undetectable rather than absent. A structured AI screen scored against a documented rubric can be audited, and the impact ratios can be published. That auditability is the strongest argument for the format, and it only holds if you actually run the audit.

This guide is general information about US employment law and AI hiring rules as of July 2026, not legal advice. Employment law changes and application depends on your facts, so confirm your obligations with counsel before deploying an AI screening tool.