Policy
The EEOC Pulled Its AI Guidance. Title VII Did Not Change.
Removing guidance does not repeal a statute. The EEOC's two AI technical assistance documents, the ADA one from May 2022 and the Title VII adverse-impact one from May 2023, came off eeoc.gov in late January 2025 and are still gone. Title VII, the ADEA, the ADA and the Uniform Guidelines all remain in force. What changed is who enforces them: agencies were directed in April 2025 to deprioritise disparate-impact enforcement, so plan for a private plaintiff.
The takeBoth readings in circulation are wrong, and the confident one is worse. Nobody who deleted a web page deleted a cause of action, and nobody saying that nothing at all changed is watching a docket. The honest version is duller: the same rules, a quieter regulator, a busier plaintiffs' bar. If this year's policy update consisted of stripping out the EEOC links and stopping there, the document got shorter without getting any easier to defend.
Where Olive fits
Open a role and see what the work shows
A private plaintiff asks what evidence stood behind the decision, which is a harder question than what the tool concluded. Olive returns six findings written by a human reviewer, each anchored to a timestamped excerpt from the candidate's own session, and the candidate gets the same report, free.
Rank your shortlistWhat came down, and when?
Two documents, both in late January 2025. The EEOC's ADA technical assistance issued May 12, 2022 and its Title VII adverse-impact technical assistance issued May 18, 2023 were taken off eeoc.gov and have not returned. Archive captures put the Title VII page at HTTP 200 through January 25, 2025 and HTTP 404 from January 31 onward; the ADA page's captures show 200 through January 26 and 404 from January 27 1.
Three details matter more than the removal itself.
The agency's own "Artificial Intelligence and the ADA" resources page still loads, and still links to the dead documents. So a hiring policy that footnotes a live eeoc.gov URL is footnoting a 404, and anyone checking the citation finds that out in one click.
Neither document ever had the force of law. Each carried a disclaimer saying so. They were the agency's reading of statutes that Congress passed and courts interpret, which is why their removal changes the reading available to you and not the standard applied to you.
No statement of reasons has surfaced. The capture dates bound the removal to a few days in January 2025 without pinning it, and nothing in the record says the agency changed its view of Title VII. Treat the documents as what the EEOC said in 2022 and 2023, cite the dated archive capture rather than the dead link, and stop describing them as current guidance.
Why the statute outlives the guidance
Because the law was never in the guidance. Disparate impact is codified at 42 U.S.C. 2000e-2(k): a plaintiff shows a particular practice causing the impact, and the employer must then show the practice is job related for the position in question and consistent with business necessity 3. The Uniform Guidelines still define a selection procedure 4, and the ADA's test-administration rule still governs how a test is given 5.
That last pair is doing more work than most policies notice. The Uniform Guidelines define a selection procedure as any measure or combination of measures used as a basis for an employment decision, and say the term covers the full range of techniques through "informal or casual interviews and unscored application forms" 4. Swapping a scored assessment for a friendly conversation does not move a hiring step outside federal selection law. It moves it somewhere with no rubric and no record.
The ADA regulation at 29 C.F.R. 1630.11 makes it unlawful to fail to administer a test in the most effective manner to ensure the result reflects the skill being measured rather than a disability that impairs sensory, manual or speaking skills, unless those skills are what the test measures 5. That is live regulation, unaffected by any removal, and it applies to a timed or interactive AI exercise exactly as it applies to a paper test. If you want the shape of the wider map, which AI hiring laws actually apply now sorts the state layer that sits on top of this one. Nothing the EEOC took off its site touched that layer: New York City's audit-and-notice duties stand 7, and Illinois has applied an effects standard since January 1, 2026 8.
Plan for a plaintiff, not an agency
Assume the next challenge arrives as a lawsuit rather than an agency charge. Executive Order 14281, signed April 23, 2025 and published at 90 Fed. Reg. 17537, directs agencies to deprioritise enforcement of every statute carrying disparate-impact liability, naming 42 U.S.C. 2000e-2 2. An order directs the executive branch and does nothing more: the private right of action, state law and state regulators are all untouched.
The named case is Mobley v. Workday, No. 3:23-cv-00770 in the Northern District of California, filed February 21, 2023. A July 2024 order allowed disparate-impact claims to proceed against the software vendor on an agent theory, and on May 16, 2025 the court granted preliminary collective certification of the ADEA age claim, finding the plaintiff had plausibly shown that all proposed collective members were subject to a common, unified policy with an alleged disparate impact 6.
Read that carefully, because it is routinely oversold. Nothing there is a finding that anyone discriminated. Preliminary certification under the ADEA is a low bar that permits notice to opt-ins and nothing else, decertification stays available, and the case is still running. What it does establish is the shape of the risk: an age claim, brought privately, aimed at a tool used across many employers at once. That is the exposure a change in federal enforcement priorities does not reduce by a single day.
Replace the dead citations in your policy
Open the policy, find every eeoc.gov reference about AI, and decide for each one whether the claim it supported still stands. Most do, because the removed documents were the agency's reading of statutes still in force. What has to change is the citation itself, since a footnote that now returns a 404 reads to a candidate, and to counsel, as a document nobody has maintained since 2023.
Four repairs, in the order they pay off:
1. Re-point each dead footnote at the dated Internet Archive capture, and say in the text that the document was withdrawn in January 2025. A withdrawn source honestly labelled is stronger than a live-looking link that fails. 2. Move the load-bearing claims onto primary sources. The statute, the Uniform Guidelines and the ADA regulation say most of what the technical assistance said, and none of them can be taken down by a change of administration. 3. Keep the vendor point, and keep its caveat. In the withdrawn Title VII document the EEOC said an employer may be responsible for a selection procedure it administers even where an outside vendor built it, and may be responsible for the acts of agents including software vendors 1. That is a 2023 agency reading rather than a rule, and it is exactly why what to get from an assessment vendor before an inquiry is worth settling in the contract instead of after a complaint. 4. Write down what you would show if asked. Defending an AI-skills assessment to legal is mostly an exercise in having the evidence already assembled, and the assembly is the slow part.
Confirm the rewrite with counsel. Public legal fact is what this page carries; what it means for your process is a question for someone who can read your contracts.
Common questions
Can the withdrawn EEOC documents still be used as a reference?
Yes, if they are labelled honestly. They read as the agency's 2022 and 2023 interpretation of the ADA and Title VII, they never carried the force of law, and both are preserved in dated Internet Archive captures. What is no longer defensible is citing them as current federal guidance or linking to a eeoc.gov URL that returns a 404. Quote them as what the EEOC said on a date, and put the statute or the regulation behind any claim the policy actually leans on.
If a vendor built the screening tool, is the liability theirs?
Not exclusively, and possibly not at all. The withdrawn Title VII document said an employer administering a selection procedure may be responsible even where an outside vendor developed it, and may be responsible for the acts of agents including software vendors given authority to act for it. That was the agency's reading rather than a rule, and it has not been tested as such in a decided case. Whether the vendor also carries exposure is the separate question being litigated in Mobley v. Workday.
Does the executive order affect New York City or Illinois obligations?
No. Executive Order 14281 directs federal agencies on enforcement priorities. It does not amend Title VII, and it leaves state and local law where it stood: New York City Local Law 144 still requires a bias audit and candidate notice 7, and the Illinois Human Rights Act amendments took effect January 1, 2026 8. The order does ask the Attorney General to consider whether federal law preempts some state rules, which is a study rather than a preemption. Nothing in it lets an employer drop a city or state duty this year.
Does an unstructured interview sit outside federal selection law?
No. The 1978 Uniform Guidelines reach anything used as a basis for an employment decision, and name informal or casual interviews and unscored application forms among the techniques covered. Validation is owed only where adverse impact appears, so a process showing no impact owes none. But the Guidelines govern the step whether or not anyone wrote anything down, which is why an undocumented conversation is the version of the step you would find hardest to defend.
What is the shortest defensible replacement for a dead EEOC footnote?
The statute or regulation the footnote was standing in for, plus a dated archive link if the agency's phrasing is doing real work. For adverse impact that means 42 U.S.C. 2000e-2(k). For what counts as a selection procedure it means 29 C.F.R. Part 1607. For how a test must be administered to a candidate with a disability it means 29 C.F.R. 1630.11. Three primary sources cover most of what a hiring policy cited the technical assistance for.
References
- 1. Select Issues: Assessing Adverse Impact in Software, Algorithms, and Artificial Intelligence Used in Employment Selection Procedures Under Title VII of the Civil Rights Act of 1964 (archived capture, 2025-01-25) web.archive.org Supports both the removal timeline for the EEOC's AI technical assistance and its 2023 answer that an employer may be responsible for a vendor-built selection procedure.
- 2. Executive Order 14281 of April 23, 2025 - Restoring Equality of Opportunity and Meritocracy govinfo.gov Supports the claim that agencies were directed to deprioritise disparate-impact enforcement, and that the order reaches the executive branch only.
- 3. 42 U.S.C. 2000e-2(k) - Burden of proof in disparate impact cases uscode.house.gov Supports the statutory disparate-impact test that survives the removal of the guidance.
- 4. 29 CFR Part 1607 - Uniform Guidelines on Employee Selection Procedures (1978), sections 1607.16(Q) and 1607.3(A) govinfo.gov Supports the definition of a selection procedure, including informal or casual interviews and unscored application forms.
- 5. 29 CFR 1630.11 - Administration of tests (Regulations to Implement the Equal Employment Provisions of the Americans with Disabilities Act) govinfo.gov Supports the live ADA duty on how employment tests are selected and administered, unaffected by the guidance removal.
- 6. Mobley v. Workday, Inc., 3:23-cv-00770 (N.D. Cal.) clearinghouse.net Supports the claim that the live AI hiring litigation is private, and the May 2025 preliminary collective certification of the ADEA claim.
- 7. Automated Employment Decision Tools: Frequently Asked Questions nyc.gov Supports the New York City bias-audit and candidate-notice duties that the federal enforcement shift leaves in place.
- 8. HB3773 Enrolled (Public Act 103-0804), amending the Illinois Human Rights Act ilga.gov Supports the January 1, 2026 effective date of the Illinois Human Rights Act amendments on AI in employment decisions.
8 sources, numbered by first appearance. How Olive sources claims
General guidance, not legal advice. Hiring rules differ by state and country and change often; check anything here against your own counsel before you act on it.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.