Can my AI product screen job candidates or evaluate workers?
The rules for AI that screens resumes, interviews candidates, or evaluates employees, including NYC's bias audit law, Illinois and California rules, discrimination law, and why vendors as well as employers are being sued.
Short answer
Yes, but employment is one of the most regulated uses of AI. Discrimination laws apply to hiring decisions whether a person or a tool makes them, and a growing set of state and city laws add bias audits, notices, and record-keeping. The companies that build these tools are also being sued directly, not just the employers that use them. Building in testing, notices, human review, and records from the start makes the product easier to sell to employers, who will ask how it helps them comply.
What the rules cover
Discrimination law applies to tools
- Federal law. Title VII, the Age Discrimination in Employment Act, and the Americans with Disabilities Act apply to decisions made with AI. A tool that screens out a protected group at a higher rate can create liability even if no one intended it, and assessments need a path for disability accommodations.
- Vendors can be liable too. In Mobley v. Workday, a federal court in California allowed discrimination claims against the maker of AI screening software to proceed on the theory that it acted as an agent of the employers using it. The court let an age discrimination claim proceed as a nationwide collective action in May 2025, and in June 2026 declined to dismiss claims under California law. The case is ongoing, and Workday denies that its tools make hiring decisions.
City and state laws
- New York City. Local Law 144 requires employers and employment agencies using an automated employment decision tool for candidates or employees in the city to have an independent bias audit within a year before use, publish a summary of the results, and notify candidates in advance. A December 2025 audit by the New York State Comptroller found gaps in the city's enforcement, which may lead to more active enforcement.
- Illinois. Since January 1, 2026, the Illinois Human Rights Act prohibits employers from using AI in ways that have a discriminatory effect, including using zip codes as a proxy for protected classes, and requires notice to workers when AI is used in employment decisions. Illinois's AI Video Interview Act separately requires notice, an explanation of how the AI works, and consent before AI analyzes video interviews, plus limits on sharing and deletion on request.
- California. Civil Rights Council regulations in effect since October 1, 2025 apply California's employment discrimination law to automated decision systems. They make clear that an employer can be responsible for discrimination by a tool, including one run by a vendor acting on its behalf, treat anti-bias testing as relevant evidence, and require keeping automated decision data for four years. California's privacy rules on automated decisionmaking, which cover employment decisions, start January 1, 2027. See Do California's automated decision rules apply to my AI product?
- Colorado. Colorado's automated decision-making law, taking effect January 1, 2027, covers technology that materially influences employment decisions. See Does Colorado's automated decision-making law apply to my product?
- The EU. The EU AI Act treats AI used for recruiting, screening, and evaluating workers as high-risk, with its own obligations for providers and employers.
Background information
If your product gathers information about candidates from outside sources, such as social media, public records, or other databases, and employers use it in hiring, it may be treated as a consumer report under the Fair Credit Reporting Act, which brings accuracy, notice, and dispute requirements.
Building it well
- Test for bias before launch and on a schedule, by sex, race and ethnicity, and age where you can, and keep the results. Employers in New York City will need audit data from you.
- Avoid proxies. Review inputs such as zip codes, graduation years, gaps in employment, and names, which can stand in for protected traits.
- Keep a human in the decision. Design scores and rankings to support a recruiter's decision, and make it easy to review and override them.
- Support notices and consent. Give employers what they need to notify candidates, explain the tool, and offer alternatives or accommodations.
- Keep records of inputs, scores, model versions, and decisions for at least four years, the period California requires.
- Describe the tool accurately. Claims like "bias-free" or "objective" create risk if testing doesn't support them. See What can I claim about my AI product in marketing?
- Set the contract terms. Your customer agreements should say who runs audits, who gives notices, and how responsibility is shared.
Checklist an agent can run against the codebase
- Find every feature that scores, ranks, filters, or recommends candidates or employees.
- List the inputs each one uses, and flag possible proxies such as zip code, age-related dates, names, and photos.
- Find any bias testing code or reports, and when they last ran.
- Check whether a person reviews outcomes before candidates are rejected, and whether overrides are recorded.
- Check for video, voice, or facial analysis of candidates, and the notice and consent flows around it.
- Check whether the product pulls candidate information from outside sources.
- Check what is logged about each decision, and the retention period for those records.
- Find where candidates can request an alternative process or accommodation.
Sources
- NYC Department of Consumer and Worker Protection: Automated Employment Decision Tools
- New York State Comptroller: Enforcement of Local Law 144 (December 2025)
If this applies to you
An AI feature review covers a hiring or workforce feature before launch: which laws apply, the testing, notice, and records the product needs to support, how to describe it to customers, and the contract terms that divide responsibility with employers. For one change, such as adding video interview analysis or a new scoring input, a pre-build feature check covers that change before it ships.