EEOC Hiring Guidelines and AI: Keeping Automated Screening Fair

The EEOC hiring guidelines that have governed selection decisions for decades apply just as fully when artificial intelligence is involved. Using an algorithm to screen candidates does not change an employer's responsibilities under federal anti-discrimination law. If a tool produces unfair outcomes, the employer using it is on the hook, no matter how the tool reached its result. This guide covers what recruiters need to know to keep automated screening fair: adverse impact, disability accommodation, transparency, and the human role.

Last updated July 2026. Important update: on January 27, 2025 the EEOC removed the AI technical assistance documents it had published in May 2023 from its website, following Executive Order 14179. The guidance is gone. The underlying law is not: Title VII, the ADA and the ADEA are statutes, they remain fully in force, and disparate impact liability still applies to automated screening. What changed is that the federal government stopped explaining how it will apply the law, while states began writing their own rules. Our map of AI hiring laws by state covers where the binding requirements now sit.

Existing law still applies

There is no separate, lighter standard for AI hiring tools. Title VII, the Americans with Disabilities Act, and the Age Discrimination in Employment Act all apply to automated screening exactly as they apply to a human reviewer. The practical implication is that you cannot outsource fairness to a vendor or an algorithm. A tool that screens out protected groups at higher rates, or that disadvantages candidates with disabilities, exposes the employer to liability regardless of intent. Withdrawn guidance does not repeal a statute, and the employer is still expected to understand and monitor the tools it uses rather than treat them as a black box.

Adverse impact and the four-fifths rule

The central fairness concept is adverse impact, also called disparate impact: a neutral-seeming criterion that nonetheless filters one protected group at a substantially higher rate than another. The long-standing screening signal is the four-fifths rule, which flags a potential problem when the selection rate for one group is less than eighty percent of the rate for the most-selected group. It is a starting point for investigation, not a hard legal threshold, but it is the kind of measure that bias audits and the EEOC both look at, and it is the same impact-ratio math a NYC Local Law 144 bias audit runs on your tool. When a tool trips this signal, the question becomes whether the criterion driving it is genuinely job-related and consistent with business necessity. If it is not, it needs to change.

• Monitor selection rates by group. You cannot manage adverse impact you never measure.
• Validate your criteria. Each must-have should predict success in the role, not stand in as a proxy for something it should not. Writing them down first is what makes them reviewable, which is the point of a documented set of resume screening criteria.
• Watch for proxies. Criteria like ZIP code, gaps in employment, or specific schools can correlate with protected characteristics.

Disability accommodation

The ADA deserves specific attention because automated tools can create barriers that are easy to overlook. A screening process that disadvantages a candidate because of a disability, or that does not offer a reasonable alternative, can violate the law even when no bias was intended. Candidates must be able to request accommodations, and the process has to be able to handle them. This is one reason fully automated rejection is risky: a person who needed an accommodation, or whose circumstances the tool did not anticipate, should reach a human rather than being silently filtered out.

The EEOC's position is simple to state and demanding to meet: the employer is responsible for the fairness of every tool it uses to screen people.

Transparency and the human in the loop

Two practices do the most to keep automated screening defensible. The first is transparency: being able to explain why a candidate was ranked or screened the way they were, in terms of job-related criteria rather than an opaque score. The second is keeping a human in the loop, so that the tool informs the decision rather than making it. Both protect candidates and protect the employer, because a documented, explainable, human-reviewed process is far easier to stand behind than an automated verdict no one can account for.

How fair-by-design screening helps

This is the philosophy Prescreener is built on. It is screening-first candidate screening software that reads the inbound resume pile, applies your criteria, scores role-fit, and ranks candidates, and it is bias-audited to support EEOC and NYC Local Law 144. Every ranking comes with the reasons behind it, so the process is transparent and explainable rather than a black box, and AI use is disclosed to candidates.

Most importantly, Prescreener ranks and flags top matches for recruiter review rather than auto-rejecting anyone. The software narrows and orders the pile, but a human always makes the final hiring decision, and edge cases, including accommodation requests, reach a person. That design keeps the employer in control of fairness, which is exactly where the EEOC expects it to be.

Keeping AI hiring fair is not a matter of finding a loophole, it is a matter of applying the standards you already know to a new kind of tool. Monitor for adverse impact, validate your criteria, handle accommodations, stay transparent, and keep a human making the call. Choose screening software that is built around those same principles, and compliance stops being a constraint and becomes simply how good hiring is done.