Using AI to Screen Resumes: Benefits and Legal Risks

Post a remote job today and you might get 400 applications by Friday, many of them polished with AI writing tools. No small team can read all of them carefully. That’s why AI resume screening has moved from big-company recruiting departments into everyday applicant tracking systems that small businesses use.

The time savings are real. So is the legal exposure. If a screening tool quietly filters out older applicants, women or people with disabilities, the employer, not just the software vendor, can be on the hook. This guide explains how AI screening works, where it helps, which US laws apply as of October 2026 and how to use it without creating a discrimination problem.

This is not legal advice. Hiring laws vary by state and city and are changing quickly, so check current rules with employment counsel.

How AI resume screening works

Most tools fall into 3 groups, and many applicant tracking systems now include all of them:

• Keyword and rules filters. The older approach: reject anyone missing a required license or keyword, or who answers “no” to a knock-out question.
• Matching and ranking models. Machine learning compares each resume to the job description, or to profiles of past successful hires, and produces a score or ranked list.
• Generative AI summaries. Large language models read resumes and write short summaries, pull out skills or suggest interview questions.

The higher the tool sits on that list, the easier it is to explain. A knock-out question about a required commercial driver’s license is transparent. A model score that says one candidate is a “92% match” and another is a “61% match” usually isn’t.

The benefits: where AI screening helps

Used carefully, AI screening can make hiring faster and in some ways fairer:

• Speed. Sorting a large applicant pool in minutes instead of days, so strong candidates don’t take other offers while they wait.
• Consistency. Every resume is evaluated against the same criteria, rather than whatever a tired recruiter notices at 5 p.m.
• Skills focus. Well-designed tools can surface candidates with the right skills from non-traditional backgrounds that a quick human skim might miss.
• Better candidate experience. Faster responses and status updates instead of silence.
• Recruiter time. Less time on sorting, more on interviews and selling the role.

The risks: how screening tools go wrong

AI models learn from data. If a model is trained on your past hires, and your past hires skew toward one group, the model can learn to prefer that group. It can also pick up proxies: features that look neutral but track protected traits.

Other risks include tools that penalize unusual resume formats used by people with disabilities or screen readers, scoring that can’t be explained to a rejected candidate, and generative AI summaries that occasionally invent or drop details.

Key takeaway: Using a vendor’s AI does not move legal responsibility to the vendor. If the tool screens people out unfairly, the employer can still be liable for discrimination.

Federal law: old rules, new tools

There’s no federal law written specifically for AI hiring tools, but existing anti-discrimination laws apply whether a person or software makes the decision. Title VII of the Civil Rights Act, the Age Discrimination in Employment Act and the Americans with Disabilities Act all cover hiring.

The key concept is disparate impact: a practice that looks neutral but disproportionately screens out a protected group can be unlawful unless it’s job-related and consistent with business necessity. The federal Uniform Guidelines on Employee Selection Procedures give a rule of thumb known as the four-fifths rule. If one group’s selection rate is less than 80% of the highest group’s rate, federal enforcement agencies generally treat that as evidence of adverse impact.

For example, if 50% of male applicants pass your AI screen but only 30% of female applicants do, the ratio is 30 ÷ 50 = 60%, well under 80%. That’s a red flag worth investigating. The guidelines also note that smaller differences can matter when they’re statistically significant, so the rule is a starting point, not a safe harbor.

Courts are also testing whether vendors can share liability. In Mobley v. Workday, a federal court in California granted preliminary certification of a nationwide age discrimination collective action over AI screening in May 2025. The case had not reached a final ruling at the time of writing, but it shows that both employers and vendors are being watched.

State and city laws on AI hiring

A growing set of state and local laws adds specific duties. Here are the main ones employers ask about, as of October 2026. Our roundup of US state AI laws tracks the wider picture.

New York City Local Law 144

If you use an automated employment decision tool to screen candidates for jobs in NYC, the city’s Department of Consumer and Worker Protection requires an independent bias audit within 1 year before use, a public summary of the results and notice to candidates at least 10 business days before the tool is used.

Illinois

Illinois amended its Human Rights Act through HB 3773, signed in August 2024 and effective January 1, 2026. It covers any employer with at least 1 employee in Illinois and applies across recruitment, hiring, promotion, discipline and discharge. Employers can’t use AI in ways that have a discriminatory effect on protected classes, can’t use ZIP codes as a proxy and must give notice when AI is used.

California

California’s Civil Rights Council regulations, effective October 1, 2025, confirm that the state’s anti-discrimination law applies to automated-decision systems used in hiring and require employers to keep employment records, including automated-decision data, for at least 4 years.

Colorado

Colorado’s original AI Act never took effect as written. The legislature replaced it with SB 26-189, signed May 14, 2026, which covers automated decision-making technology in consequential decisions, including employment. It requires notice, explanations after adverse outcomes, data correction and human review on request, with key duties applying from January 1, 2027.

If you hire in Europe, the EU AI Act treats AI used in recruitment as high-risk. Our guide to the EU AI Act for US companies explains what that means.

Keeping a human in the decision

Every law and guideline above points the same way: AI can help sort, but a person should decide. In practice, “human in the loop” means more than a recruiter clicking “approve” on whatever the tool ranks first.

• Use AI to prioritize, not reject. Let the tool order the queue, but have someone review candidates below the cutoff, at least on a sample basis.
• Give reviewers authority to override the tool, and record why when they do.
• Avoid fully automated rejections except for clear, objective requirements like a legally required license.
• Offer an alternative process. Tell applicants how to request an accommodation or a non-automated review. This matters under the ADA.
• Document decisions so you can explain later why a candidate did or didn’t advance.

Where to start: a 7-step plan for employers

1. Find out what’s already on. Many applicant tracking systems have added AI scoring or matching features. Check your settings and ask your vendor.
2. Map where you hire. Note candidates in NYC, Illinois, California, Colorado and the EU, since each has specific rules.
3. Question your vendor. Ask what data trained the model, which factors drive scores, what bias testing they’ve done, whether they can support a Local Law 144 audit and how they handle accommodations.
4. Clean up your criteria. Remove requirements that act as proxies, such as graduation year limits, and make sure every criterion is truly job-related.
5. Test your own results. Compare selection rates by sex, race and ethnicity, and age where you have data, using the four-fifths rule as a first check. Repeat at least yearly and after major changes.
6. Give clear notice to candidates wherever you use AI screening, even where not legally required.
7. Write it down. Add AI hiring rules to your AI acceptable use policy and your AI governance checklist, with a named owner.

For the wider picture of AI across recruiting and onboarding, see our guide on how HR teams use AI, and for how bias gets into models in the first place, read our explainer on AI bias.

Conclusion

AI resume screening can save small teams hours on every opening and bring more consistency to early-stage hiring. But the law treats an AI screen like any other hiring practice: if it screens out protected groups without a solid job-related reason, the employer is responsible. Use AI to organize and prioritize, test the results for bias, give candidates notice and keep a trained person accountable for every hiring decision.

Sources

• NYC Department of Consumer and Worker Protection: Automated Employment Decision Tools — Local Law 144 bias audit, publication and 10-business-day notice rules, enforced from July 5, 2023
• 29 CFR 1607.4, Uniform Guidelines on Employee Selection Procedures (Cornell LII) — the four-fifths rule for adverse impact
• California Civil Rights Council: regulations on automated-decision systems — effective October 1, 2025, 4-year record retention
• Colorado General Assembly: SB 26-189, Automated Decision-Making Technology — signed May 14, 2026; notice, adverse-outcome disclosure, correction and human review duties