A company deployed AI to screen resumes.
The AI ranked 5,000 applicants and eliminated 4,500 before any human reviewed them.
It was fast, efficient, and seemed objective.
Then someone analyzed what data the AI actually used to make decisions.
ZIP codes. College graduation years. Employment gaps. Names that suggested certain ethnicities.
The AI was using proxy data that correlates with race, age, disability, and gender to filter candidates.
Every screening decision violated EEOC guidelines.
Now the company faces a class action from thousands of rejected applicants.
How Companies Use AI for Recruiting
AI recruiting tools are everywhere because they save time, screen thousands of resumes instantly, rank candidates objectively, and reduce human bias in initial screening.
Resume screening: AI reads resumes and identifies qualified candidates.
Candidate ranking: AI scores applicants and presents top candidates to hiring managers.
Interview scheduling: AI determines which candidates advance to interviews.
Assessment analysis: AI evaluates skills tests and personality assessments.
Most companies deployed AI recruiting without understanding what data the AI actually uses to make decisions.
The Proxy Discrimination Problem
AI doesn’t need to see race, age, or gender to discriminate based on those factors.
It uses proxy data that correlates with protected characteristics.
ZIP codes proxy for race: Certain ZIP codes are predominantly minority. AI learns applicants from those ZIP codes are “lower quality” based on biased training data.
College graduation year proxies for age: AI uses graduation dates to infer age and deprioritizes older candidates.
Employment gaps proxy for disability and caregiving: AI penalizes gaps that correlate with medical leave or parental responsibilities.
Name analysis proxies for ethnicity: AI infers ethnicity from names and treats certain names differently.
School prestige proxies for socioeconomic status: AI favors candidates from expensive schools, which correlates with race and economic background.
The AI never sees “race” or “age” as data fields. But it uses information that produces the same discriminatory outcomes.
Why This Violates EEOC Guidelines
The EEOC is clear: if AI produces discriminatory outcomes, it violates Title VII regardless of intent.
Disparate impact: If AI screening disproportionately eliminates protected groups, it’s illegal even without intentional discrimination.
Validation required: If AI adversely affects protected groups, the company must prove the AI criteria are job-related and consistent with business necessity.
Can’t validate proxy factors: Companies can’t prove ZIP codes predict job performance. They can’t validate employment gaps as legitimate screening criteria.
Most companies using AI recruiting have never tested for adverse impact and couldn’t validate their AI’s criteria if challenged.
The FCRA Complication
The Fair Credit Reporting Act applies to AI recruiting in ways companies don’t expect.
If AI uses data beyond the resume (social media, public records, credit history), it may be functioning as a background check requiring FCRA compliance including disclosure to candidates, consent before screening, and adverse action notices when AI rejects candidates.
Most AI recruiting tools don’t provide FCRA-compliant adverse action notices because companies don’t realize FCRA applies.
Real Scenarios
The ZIP code filter: A retail company used AI to screen cashier applicants. The AI eliminated candidates from certain ZIP codes because past employees from those areas had higher turnover. The ZIP codes were predominantly minority neighborhoods. EEOC found discrimination affecting thousands of applicants.
The college degree proxy: A tech company’s AI prioritized candidates from top-tier universities. This disproportionately excluded minority candidates who attended less prestigious schools. The degree requirement wasn’t actually necessary for the jobs. Class action followed.
The name bias: An AI recruiting system learned to deprioritize candidates with names associated with certain ethnicities based on biased training data from past hiring decisions. The pattern was discovered during EEOC investigation.
What Companies Should Do
If AI is screening candidates, testing and validation are mandatory.
Adverse impact testing: Analyze whether AI disproportionately rejects protected groups. Run this test before deployment and regularly thereafter.
Review AI criteria: Understand what data the AI actually uses. Eliminate proxy factors like ZIP codes, graduation years, and employment gaps.
Validate job-relatedness: Prove every AI screening criterion is actually predictive of job performance and necessary for the role.
FCRA compliance: If AI uses data beyond the resume, implement FCRA adverse action notices.
Human oversight: Don’t let AI eliminate candidates without human review of the criteria and outcomes.
Documentation: Keep records of AI decisions, testing results, and validation studies.
Most companies skipped all of this because AI recruiting “seemed objective.”
The Question to Ask
Does AI screen job candidates before humans see them?
If yes, what data does the AI actually use to make decisions? Has anyone tested whether it discriminates against protected groups? Can the company prove that the AI’s criteria predict job performance?
If nobody knows the answers, there’s an EEOC complaint waiting to happen.
Schedule a 30-minute AI Recruiting Compliance Assessment where we’ll:
- Review what data AI recruiting tools actually use
- Test for adverse impact on protected groups
- Identify proxy discrimination in AI criteria
- Show what FCRA compliance requires
- Create validation studies for AI screening factors
Because AI recruiting feels efficient until the EEOC investigation reveals it rejected 4,500 people using ZIP codes as a ranking factor.
Brian Alexander
Founding Partner, VisionAI+ Consulting Group
Former General Counsel, AI Engineering Firm
Certified ISO 42001 Lead Auditor