EEOC Guidance on AI in Hiring: What to Know
The EEOC's guidance on AI in hiring establishes that employers stay legally responsible under federal anti-discrimination law when they use automated employment decision tools, including resume screeners, chatbots, and video interview scorers. Under Title VII of the Civil Rights Act, the Americans with Disabilities Act (ADA), and the Age Discrimination in Employment Act (ADEA), an AI hiring tool that produces a disparate impact on a protected group can expose an employer to liability even when a third-party vendor built the tool. The agency's position is direct: compliance cannot be outsourced to software.
This article explains what the EEOC has said, which laws apply, what the agency expects employers to do, and how this fits with state and local AI hiring rules.
What is the EEOC's guidance on AI in hiring?
The U.S. Equal Employment Opportunity Commission (EEOC) is the federal agency that enforces workplace anti-discrimination laws. Beginning in 2021, the EEOC launched an initiative to examine how artificial intelligence, machine learning, and algorithmic decision-making affect hiring, promotion, and other employment decisions. It has since issued technical assistance documents interpreting how existing statutes apply to these tools.
The core message across the EEOC's published materials is that existing civil rights law already covers AI hiring tools. No new statute was required for the agency to act. If a software system screens, ranks, or rejects candidates, the legal standards that apply to a human recruiter apply to the algorithm.
The two most cited EEOC technical assistance documents address:
The Americans with Disabilities Act (ADA) and how algorithmic tools can illegally screen out candidates with disabilities.
Title VII and how selection procedures built on AI can create a disparate impact based on race, color, religion, sex, or national origin.
Which laws does the EEOC apply to AI hiring tools?
The EEOC enforces several federal statutes, and each one reaches AI tools used in employment decisions:
Law: Title VII (1964)
What it prohibits: Employment discrimination based on race, color, religion, sex, or national origin.
How it applies to AI hiring: An AI scoring or ranking tool that disproportionately rejects candidates from a protected group may be considered an unlawful selection procedure.
Law: Americans with Disabilities Act (ADA) (1990)
What it prohibits: Discrimination against qualified individuals with disabilities.
How it applies to AI hiring: An AI tool that screens out candidates who need reasonable accommodations or evaluates traits unrelated to job performance may violate the ADA.
Law: Age Discrimination in Employment Act (ADEA) (1967)
What it prohibits: Employment discrimination against individuals aged 40 and older.
How it applies to AI hiring: An AI algorithm that directly or indirectly down-ranks older applicants, including through age-related proxies, may violate the ADEA.
The EEOC's guidance ties these tools to the Uniform Guidelines on Employee Selection Procedures (UGESP), the 1978 framework that defines how to measure adverse impact. The widely referenced four-fifths rule (the selection rate for a protected group should be at least 80% of the rate for the highest-selected group) comes from this framework and applies to AI-driven selection the same way it applies to a written test.
What counts as an automated employment decision tool?
The EEOC and parallel state and local laws use broad definitions. An automated employment decision tool generally includes any computational process that issues a score, classification, recommendation, or ranking used to make or substantially assist an employment decision. Common examples:
Resume screeners that filter or rank applicants by keyword, experience, or inferred fit.
Chatbots that conduct initial screening conversations and route candidates.
Video interview analysis that scores facial expression, word choice, or tone.
Game-based assessments that infer personality or cognitive traits.
Matching engines that surface or hide candidates from recruiter view.
The legal exposure does not depend on whether a human makes the final call. If the tool narrows the pool, it has influenced the decision.
How can AI hiring tools cause illegal discrimination?
AI tools produce discrimination in two main ways the EEOC has identified: disparate treatment (intentional differential treatment) and, far more commonly, disparate impact (a neutral-seeming process that disadvantages a protected group). Most AI hiring liability falls under disparate impact, because the model is rarely told to discriminate. It learns patterns from historical data, and that data carries the results of past human decisions.
Why does training data create bias?
Machine learning models predict by finding patterns in examples. If a company's past hires skewed toward one group, a model trained to find "successful" candidates can learn the attributes correlated with that group and reproduce the skew.
The most documented example is Amazon's internal AI recruiting tool, which the company scrapped around 2018 after engineers found it down-ranked resumes that included signals associated with women, such as the word "women's" or certain all-women colleges. The model had been trained on roughly a decade of resumes submitted to a male-dominated technical workforce, so it learned to prefer patterns common among men. Amazon never used it to evaluate candidates in production, but the case became a reference point for how training data encodes historical bias.
How do proxy variables work?
A model does not need a protected characteristic as an input to discriminate by it. It can use proxy variables, which are neutral data points that correlate with a protected class. Examples:
ZIP code can correlate with race because of residential segregation.
Years since graduation or graduation date can serve as a proxy for age.
Gaps in employment history can correlate with sex (caregiving) or disability.
Names, hobbies, or affiliations can correlate with national origin, religion, or sex.
Because these proxies can sit inside a model that holds thousands of features, bias can operate without any human intending it and without an obvious discriminatory variable on the form.
What enforcement actions show how this works in practice?
Two cases show how AI hiring liability turns into real consequences.
The iTutorGroup EEOC settlement (2023). The EEOC alleged that the tutoring company's application software was configured to automatically reject female applicants age 55 and older and male applicants age 60 and older. The company settled with the EEOC for $365,000. The case is significant because it was the EEOC's first public settlement involving an AI hiring tool and showed the agency will pursue employers whose software encodes age discrimination, even when the cutoff is a simple programmed rule rather than a complex model.
Mobley v. Workday. This lawsuit alleges that AI screening software offered by the vendor Workday discriminated against applicants on the basis of age, race, and disability. A federal court allowed the case to proceed and let a claim advance on the theory that the vendor itself could be treated as an agent participating in employment decisions. The case matters because it tests whether liability can attach not only to the employer using the tool but to the company that built and operated it. The court dismissed the intentional discrimination claim while allowing the disparate impact claim to move forward. The litigation is ongoing, and outcomes may change as it develops.
These cases reinforce the EEOC's stated position: the party that makes the decision holds responsibility for it, whatever the software license says.
What does the EEOC expect employers to do?
The EEOC has not published a single mandatory checklist, but its technical assistance and enforcement posture point to clear expectations. Employers using AI hiring tools should be able to demonstrate the following.
Audit for adverse impact. Test selection rates across race, sex, age, and other protected categories using the four-fifths rule and other statistical measures. Do this before deployment and on a recurring schedule.
Document job-relatedness. Under Title VII, a tool that causes adverse impact can still be lawful if the employer shows the selection procedure is job-related and consistent with business necessity. Keep evidence linking each measured trait to the actual demands of the role.
Provide accommodations. Under the ADA, tell applicants that reasonable accommodations are available and give a clear way to request an alternative format. A timed game or a video assessment may disadvantage a candidate with a disability who could perform the job.
Vet vendors directly. Ask vendors for their own bias-audit results, the validation studies behind their tools, and the data the model was trained on. The EEOC has signaled that an employer cannot escape liability by pointing to a vendor's assurances.
Keep a human in the loop with real authority. A human reviewer who only rubber-stamps the algorithm does not reduce exposure. The reviewer needs the information and authority to override the tool.
How does this connect to state and local AI hiring laws?
Federal EEOC enforcement runs in parallel with a growing set of state and local rules. Employers operating across jurisdictions face overlapping obligations, and the strictest applicable rule controls. Key examples:
NYC Local Law 144, which requires an independent bias audit of an automated employment decision tool within the prior year, public posting of the audit summary, and advance notice to candidates. Enforcement began in July 2023.
The Illinois Artificial Intelligence Video Interview Act, effective January 2020, which requires notice and consent before using AI to analyze video interviews and limits sharing of those videos.
The Colorado AI Act (SB 24-205), which treats employment-related AI as high-risk and imposes duties on developers and deployers to guard against algorithmic discrimination.
Outside the U.S., the EU AI Act classifies AI systems used in recruitment and employment decisions as high-risk under Annex III, triggering documentation, risk-management, and human-oversight requirements for companies operating in or hiring into the EU. For a jurisdiction-by-jurisdiction breakdown of U.S. requirements, see AI hiring laws by state.
In practice, the EEOC's federal anti-discrimination standards set the minimum that applies everywhere, and whichever state or local law reaches the employer can add stricter duties on top.
Next steps: an AI hiring compliance checklist
Use this checklist to align an AI hiring program with EEOC guidance and adjacent laws.
Step: 1
Action: Inventory every automated tool used in hiring or employment decisions.
Source of obligation: EEOC guidance.
Step: 2
Action: Perform a pre-deployment adverse impact analysis using the four-fifths rule.
Source of obligation: Title VII and the Uniform Guidelines on Employee Selection Procedures (UGESP).
Step: 3
Action: Document the job-relatedness and business necessity of each AI tool.
Source of obligation: Title VII.
Step: 4
Action: Build an accommodation request process into every AI-assisted assessment.
Source of obligation: Americans with Disabilities Act (ADA).
Step: 5
Action: Verify that the model does not rely on proxies for age.
Source of obligation: Age Discrimination in Employment Act (ADEA).
Step: 6
Action: Obtain and review the vendor's bias audit and model validation documentation.
Source of obligation: EEOC guidance on employer responsibility for vendor-provided AI tools.
Step: 7
Action: Conduct an independent bias audit if the AI tool is used for hiring in New York City.
Source of obligation: NYC Local Law 144.
Step: 8
Action: Provide required notice and obtain consent when using AI video analysis in Illinois.
Source of obligation: Illinois AI Video Interview Act.
Step: 9
Action: Classify the AI system as high-risk if it falls under Colorado AI Act or EU AI Act requirements.
Source of obligation: Colorado AI Act and EU AI Act.
Step: 10
Action: Schedule recurring bias audits and retain all compliance documentation.
Source of obligation: EEOC's ongoing compliance expectations.
Frequently asked questions
Does the EEOC have a specific AI hiring regulation?
No. The EEOC has not issued a standalone AI regulation. It enforces AI hiring through existing statutes (Title VII, the ADA, and the ADEA) and has published technical assistance documents explaining how those laws apply. An earlier EEOC AI initiative and related materials were de-emphasized after a change in administration, but the underlying anti-discrimination statutes stay in full force.
Is an employer liable if a vendor's AI tool discriminates?
Generally yes. The EEOC's position is that an employer is responsible for the selection tools it uses, including those built and operated by vendors. The Mobley v. Workday litigation is testing whether a vendor can also be held liable as an agent in the hiring process, but employer responsibility for outcomes is well established.
What is the four-fifths rule for AI hiring?
The four-fifths rule is a benchmark from the Uniform Guidelines on Employee Selection Procedures. A selection process shows potential adverse impact if the selection rate for any protected group is less than 80% of the rate for the highest-selected group. It applies to AI scoring and ranking tools the same way it applies to a traditional test.
Do federal and state AI hiring laws conflict?
They generally stack rather than conflict. The EEOC sets a federal anti-discrimination baseline that applies everywhere. State and local laws such as NYC Local Law 144, the Illinois AI Video Interview Act, and the Colorado AI Act add their own audit, notice, and consent duties on top. Employers must meet whichever requirement is strictest in each jurisdiction.
