EEOC has confirmed that Title VII disparate impact doctrine applies to AI hiring tools. Employers using AI to screen, rank, or evaluate applicants bear responsibility for discriminatory outcomes — even when the tool comes from a third-party vendor. This checklist gives you the 8 steps to assess your exposure and build a defensible audit trail.
Why Existing Civil Rights Law Applies to AI Hiring Tools
Note on EEOC guidance status: EEOC issued a Technical Assistance document in May 2023 applying Title VII, ADA, and ADEA to AI hiring tools. That guidance was withdrawn in January 2025 per Executive Order 14173 ("Ending Illegal Discrimination and Restoring Merit-Based Opportunity"). The withdrawal of the guidance document does not change the underlying law: Title VII, ADA, and ADEA remain in force, and disparate impact doctrine continues to apply to AI hiring tools under the same legal framework that has governed pencil-and-paper tests since the 1970s. Employer liability for discriminatory AI screening is a function of statute and case law, not EEOC guidance.
Three federal statutes are directly relevant to AI hiring tools:
- Title VII (Civil Rights Act): prohibits employment discrimination based on race, color, religion, sex, or national origin
- ADEA (Age Discrimination in Employment Act): covers workers 40 and older
- ADA (Americans with Disabilities Act): requires employers to make reasonable accommodations — and prohibits screening out qualified candidates with disabilities
The key doctrine is disparate impact: even a facially neutral AI tool can violate these laws if it produces outcomes that disproportionately harm a protected class. Intent is irrelevant. If the outcome is discriminatory, the employer is liable.
The 8-Step EEOC AI Hiring Compliance Checklist
Step 1: Classify Which Tools Are Subject to EEOC Guidance
List all AI tools used in your hiring, promotion, or employment decision process. A tool is subject to EEOC guidance if it:
- Screens, ranks, or scores applicants or employees
- Analyzes video, audio, or text from interviews
- Predicts performance, engagement, or retention
- Influences who proceeds to the next stage or receives an offer
Tools used only for scheduling, calendar management, or administrative tasks are generally outside scope.
Step 2: Obtain Vendor Adverse Impact Studies
For each in-scope tool, request from the vendor:
- Adverse impact study for the specific tool and version you use
- Demographic groups tested (race/ethnicity, sex, age, disability status)
- Job categories and tasks the validation covers
- Selection rate data by group and the four-fifths rule calculation
- Date of the study and whether it covers your specific use case
If the vendor cannot provide this documentation, or provides only a general company-level statement rather than tool-specific validation, that is insufficient. Treat it as a gap requiring your own investigation.
Step 3: Run Your Own Four-Fifths Rule Calculation
Do not rely solely on vendor studies. After the tool has been running for at least 30 days, pull your selection data and calculate:
Four-fifths rule formula: Selection rate for protected group ÷ Selection rate for highest-selected group
If the result is below 0.80 (80%), document the disparity and investigate the cause.
Minimum sample size for statistical reliability: 30 per group. Below 30, you cannot run the calculation meaningfully — document this limitation and plan to revisit when you have sufficient data.
Step 4: Document Business Necessity
If adverse impact exists, EEOC requires you to demonstrate that the tool is justified by business necessity. Document:
- The specific job-related competency or attribute the tool measures
- Why that competency is necessary for the role
- Why the AI tool is an appropriate way to measure it
Business necessity is a high bar. "The vendor said it predicts performance" is not enough. You need evidence that the measured attributes are genuinely required for the job.
Step 5: Document Job-Relatedness and Validation Evidence
Tools must be validated for the specific job and context you use them in. Three acceptable validation approaches under EEOC's Uniform Guidelines on Employee Selection Procedures:
| Validation type | What it shows |
|---|---|
| Content validity | The tool measures skills directly required for the job |
| Criterion validity | Tool scores correlate with actual job performance |
| Construct validity | The tool measures a construct (e.g., "analytical thinking") linked to job performance |
Request the vendor's validation study. Confirm it covers your job category, not just a general population. If you are using a tool validated for one role type in a different role type, you may need a separate validation study.
Step 6: Implement Candidate Disclosure Where Required
Federal EEOC guidance does not require employers to disclose AI use to candidates. But these state and local laws do:
| Jurisdiction | Requirement | Effective |
|---|---|---|
| New York City Local Law 144 | Bias audit required; results published; candidates notified | Already in effect |
| Illinois AI Video Interview Act | Candidates notified AI analyzes their interview; written consent required | Already in effect |
| Maryland HB 1202 | Candidates notified before AI facial recognition analysis used in hiring | Already in effect |
| Connecticut SB 5 | Disclosure when AI used in consequential employment decisions | October 1, 2027 |
Check your jurisdiction and add disclosure language to your job application or interview confirmation email if required.
Step 7: Establish a Bias Monitoring Cadence
EEOC compliance is not a one-time audit. AI tools can drift — model updates, changes in applicant pools, or changes in job requirements can shift outcomes over time. Build a monitoring cadence:
- Quarterly: Pull selection rate data by protected group; recalculate four-fifths rule
- Annually: Full audit — re-request vendor validation documentation; review any EEOC guidance updates
- On tool update: Rerun adverse impact analysis after any vendor model update
Assign one person as the owner of this monitoring. Without an owner, it does not happen.
Step 8: Build Your Audit Trail
Maintain a single document per AI hiring tool with:
- Tool name, vendor, version, and date first deployed
- Scope: job roles and stages where tool is used
- Vendor validation documentation (save a copy — vendors change their documentation)
- Your four-fifths rule calculations by quarter
- Disclosure language used and jurisdictions covered
- Any adverse impact findings and actions taken
If EEOC investigates a complaint, this documentation demonstrates good faith compliance. The absence of documentation is itself a liability.
Where to Go Next
For the full picture of AI employment law in the US and EU, including the EU AI Act's Annex III requirements for AI hiring tools, see HR AI governance — what EU AI Act and EEOC require. For state law requirements across all US jurisdictions, see AI hiring tool compliance — US state laws 2026.
Before adding any new AI hiring tool, run it through the AI vendor due diligence checklist — Step 2 above (vendor documentation) maps directly to what to request in that process.
