EEOC guidance on AI in hiring: what it said, what was removed, what still applies

The EEOC published two technical assistance documents on AI in hiring, one on the Americans with Disabilities Act in May 2022 and one on Title VII adverse impact in May 2023. Both were removed from the agency's website on January 27, 2025, after the new administration revoked the previous AI executive order, and as of September 2026 their pages return "not found." The EEOC's National Enforcement Plan of June 4, 2026 does not mention AI at all and says the agency will not litigate disparate impact claims. None of that changed the statutes the guidance explained: Title VII, the ADA and the ADEA still apply to a hiring decision whether a person or software produced it.

This page sets out what the EEOC said, what it withdrew and when, its current position, and what an employer using AI tools should still do. For the disparate impact theory in depth, including the four-fifths calculation, see Title VII disparate impact and AI hiring tools.

Timeline

One oddity worth knowing if you go looking: the EEOC's "Artificial Intelligence and the ADA" resource page is still online and still lists the 2022 document and its companion tips for workers, but both links lead to "page not found." Copies of the removed documents have been preserved by third parties, including the ACLU of Massachusetts; they are archives, not current agency positions.

What the removed guidance said

Both documents stated that their contents "do not have the force and effect of law." They explained how existing law applied to algorithmic tools. The main points:

Title VII (2023)

• Vendor tools are the employer's responsibility "in many cases." An employer that administers a selection procedure may be responsible if it discriminates, "even if the test was developed by an outside vendor," and employers can be responsible for agents they authorize.
• Ask the vendor. At a minimum, whether the tool has been evaluated for substantially lower selection rates for protected groups, and whether that evaluation used the four-fifths rule or statistical significance.
• The four-fifths rule "is merely a rule of thumb" and may be inappropriate in some circumstances, such as large numbers of selections.
• Less discriminatory alternatives. Failing to adopt a less discriminatory algorithm that was considered during development could give rise to liability, and employers were encouraged to self-audit on an ongoing basis.

ADA (2022)

The document named three common ways an algorithmic tool could violate the ADA:

1. not providing a reasonable accommodation needed for the applicant to be rated fairly and accurately;
2. relying on a tool that "screens out" an individual with a disability who could do the job with a reasonable accommodation, for example video software analyzing speech patterns that scores an applicant with a speech impediment unfairly; and
3. using a tool that makes disability-related inquiries or medical examinations the ADA restricts.

It called telling applicants what an assessment involves and how to request an accommodation a "promising practice."

What the EEOC says now

The National Enforcement Plan, signed by Chair Andrea Lucas on June 4, 2026, states that "pursuant to Executive Order 14281," the EEOC "will prioritize disparate treatment theories of liability," "will eliminate the use of disparate impact liability theories in investigations 'to the maximum degree possible,'" and "will not commence, develop, or continue to pursue litigation advancing disparate impact claims." We searched the signed plan for artificial intelligence, algorithms, automated tools and technology and found no reference to any of them. The previous Strategic Enforcement Plan had named the use of technology, "including artificial intelligence and machine learning," to recruit applicants or make hiring decisions as a priority where such systems intentionally exclude or adversely impact protected groups.

Read together: the EEOC is unlikely to bring a case arguing that an AI tool had an unintended discriminatory effect. It can still pursue intentional discrimination, and the one AI-related case it settled, iTutorGroup, was of that kind: the EEOC alleged the software was programmed to reject female applicants aged 55 or older and male applicants aged 60 or older.

What still applies, with or without guidance

What employers using AI tools should do in 2026

1. Keep asking vendors the questions the guidance listed. Adverse impact testing, the method used, and what happens when a new model version ships. The questions came from the law, not the document.
2. Run your own numbers where a tool screens or scores. Private plaintiffs and state agencies can still bring disparate impact claims.
3. Make accommodations easy to request. ADA screen-out is written into the statute. Say in advance what an assessment involves and how to ask for an adjustment; see ADA accommodations in interviews.
4. Know which of your tools decide and which only record. A transcription tool that feeds a person's judgment is in a different position from one that rejects applicants automatically.
5. Keep records. Federal rules require keeping hiring records, and California requires four years for automated-decision data.
6. Watch the states. Federal retreat has not stopped state activity; several notice regimes start in 2027.

A working internal policy that covers these steps is in how to use AI in recruiting responsibly.