Using AI To Screen Job Candidates? Your Company Still Owns The Legal Risk

Buying an AI hiring tool does not hand the legal responsibility for hiring decisions to the vendor. Under US employment law, the employer that uses a selection procedure is generally answerable for its effects, whoever built it.

That point gets lost in sales demos. Tools that rank resumes, score recorded interviews, run game-based assessments or chat with applicants promise speed and consistency, and some deliver it. But any tool that filters people out of a hiring process is a selection procedure, and selection procedures have been regulated for decades.

How discrimination claims attach to software

Federal law prohibits employment discrimination based on protected characteristics including race, color, religion, sex, national origin, age and disability. Two legal theories matter most for automated tools.

The first is disparate treatment: a tool that explicitly uses a protected characteristic, or an obvious stand-in for one, to make decisions. The second, and the more likely risk with AI, is disparate impact. A practice that looks neutral can still be unlawful if it disproportionately screens out a protected group and the employer cannot show it is job-related and consistent with business necessity.

Machine-learning models learn from historical data. If past hiring favored certain schools, zip codes, career paths or speech patterns, a model can learn those preferences as signals of a good hire without anyone telling it to. The result can be disparate impact that nobody intended and nobody noticed.

The federal Uniform Guidelines on Employee Selection Procedures describe a rule of thumb, often called the four-fifths rule, for spotting possible adverse impact by comparing selection rates between groups. It is a screening heuristic rather than a safe harbor, but it is a reasonable starting point for the questions you put to a vendor and to your own data.

Disability and accommodation

Disability law creates a separate set of obligations. An assessment that measures reaction time, eye contact, speech fluency or facial expressions may disadvantage applicants with certain disabilities in ways unrelated to the job. Employers are expected to provide reasonable accommodations during hiring, which means candidates need a clear way to request an alternative format and a human who will respond. If your application flow is fully automated with no visible path to a person, fix that before anything else.

State and city laws add disclosure duties

On top of federal law, a growing number of states and cities regulate automated hiring tools directly. New York City requires employers using covered automated employment decision tools on candidates or employees in the city to obtain an independent bias audit, publish a summary of the results and give candidates advance notice. Illinois regulates the use of AI to analyze video interviews, including notice and consent requirements. Other states have passed or are considering broader laws on high-risk automated decisions.

These rules change quickly and often turn on where the candidate is located, not where your company is headquartered. A remote-first startup hiring nationally can be subject to several regimes at once. Check the current requirements for every jurisdiction you hire in, and treat telling candidates about automated tools as a default practice rather than a legal minimum.

Questions to put to a hiring-tech vendor

Ask what the tool measures, what data it was trained on and how it defines a successful candidate. If the vendor cannot explain what drives a score in plain language, you will not be able to explain it to a regulator or to a plaintiff's lawyer.

Ask for any bias audits or adverse impact analyses, who performed them, on what data and how recently. An audit run on the vendor's general population tells you less than one run on your own applicant pool.

Read what the contract says about responsibility. Many vendor agreements disclaim liability for employment outcomes, which reinforces the basic point: the risk stays with you unless you negotiate otherwise.

Find out what data the tool keeps about candidates, for how long, and whether it is used to train the vendor's models for other customers.

What to do before you switch it on

Use AI to assist human reviewers rather than to reject candidates automatically, at least until you have evidence the tool performs fairly on your own pipeline.

Run your own adverse impact analysis on outcomes at regular intervals and keep the records. Monitoring after launch matters as much as the vendor's assurances before it.

Tell candidates when automated tools are used, what they assess and how to request an accommodation or an alternative process.

Document why each assessed criterion relates to the job. If a tool scores something you cannot connect to the work, drop that feature or drop the tool.

And bring in employment counsel before rollout, not after a complaint. The cost of a review is small next to the cost of defending a selection process nobody at your company can explain.

This is general information, not legal advice. Speak to a qualified attorney in your jurisdiction before acting on any of it.

Sources

EEOC — Prohibited Employment Policies/Practices

NYC Department of Consumer and Worker Protection — Automated Employment Decision Tools