Algorithmic Bias in Hiring Tools: Legal Risks for Employers

Originally published in Spanish on derechoartificial.com. Adapted for the US audience by Elena Markov.

Introduction

The integration of artificial intelligence into employment decisions has accelerated dramatically. Resume screening algorithms, video interview analysis systems, psychometric assessments, and chatbot-based initial interviews now filter millions of job applicants annually. While these tools promise efficiency and objectivity, they also introduce significant legal liability for employers who deploy them without proper oversight.

The legal framework governing AI in hiring is evolving rapidly. Federal civil rights law, Equal Employment Opportunity Commission guidance, and an emerging patchwork of state regulations create a compliance landscape that demands proactive attention from employers, HR professionals, and legal counsel.

This analysis examines the current legal framework, identifies risk areas, and provides guidance for navigating algorithmic hiring while minimizing discrimination liability.

Title VII and Disparate Impact Theory

The Facially Neutral Standard

Title VII of the Civil Rights Act of 1964 prohibits employment discrimination based on race, color, religion, sex, and national origin. Critically, this prohibition extends to facially neutral practices that disproportionately affect protected classes—a concept known as disparate impact.

Under the Uniform Guidelines on Employee Selection Procedures (UGESP), a hiring practice has an adverse impact when it results in a selection rate for a protected group that is less than 80% of the rate for the group with the highest selection rate (the "80% rule" or "four-fifths rule").

How AI Systems Create Disparate Impact

AI hiring tools can produce discriminatory outcomes through several mechanisms:

Training data bias: Systems trained on historical hiring data may learn and perpetuate past discriminatory patterns. If successful employees from previous decades disproportionately represented certain demographic groups, the algorithm may learn to prefer similar profiles.

Proxy variable correlation: AI systems may identify variables that correlate with protected characteristics even when those characteristics are not explicitly used. Zip codes correlate with race due to historical segregation. Names correlate with national origin. Employment gaps may correlate with caregiving responsibilities that disproportionately affect women.

Feature selection: The choice of which data inputs to use in an algorithm can encode bias. Systems that analyze voice patterns, facial expressions, or communication style may disadvantage speakers of accented English, individuals with certain disabilities, or neurodivergent candidates.

Feedback loops: AI systems that learn from ongoing hiring decisions may amplify initial biases, creating self-reinforcing discriminatory patterns.

Case Law and Enforcement Actions

The EEOC has brought enforcement actions against employers for AI-driven hiring discrimination:

EEOC v. JPMorgan Chase (2013): The Commission found that JPMorgan's parolee-exclusion policy violated Title VII because it disproportionately excluded Black and Hispanic applicants without business necessity justification.

EEOC guidance on AI (2022): The Commission issued technical assistance noting that employers may be liable for AI tools that screen out protected groups, even if the employer did not develop or intend the discriminatory outcome.

State AG actions: Attorneys general in New York, California, and Illinois have investigated employers using AI hiring tools, citing disparate impact concerns.

EEOC Guidance on AI in Hiring

The 2022 Technical Assistance

In May 2022, the EEOC issued guidance clarifying that existing civil rights law applies to AI hiring tools. Key points include:

Employer responsibility: "If an employer administers a test or selection procedure, including one that is automated, it may be responsible under Title VII if the test or selection procedure discriminates on a basis prohibited by Title VII."

Vendor reliance does not eliminate liability: "An employer may be responsible [for discrimination] if the employer administers a test that is developed and administered by another party."

Disparate impact defense: Employers can defend disparate impact claims by demonstrating that the practice has a business necessity and is job-related. However, if an alternative practice with less discriminatory impact is available and the employer fails to adopt it, liability may attach.

The 2024 Guidance on AI and Automated Systems

Building on earlier guidance, the EEOC issued additional technical assistance addressing:

Assessment tools: AI systems that evaluate job candidates must be validated for job-relatedness. If a video interview AI analyzes facial expressions, this must be validated as predictive of job performance.

Algorithmic bias audits: Employers should conduct disparate impact analysis of AI tools before deployment and monitor for emerging bias during use.

Accommodation obligations: AI systems must provide reasonable accommodation for applicants with disabilities. A system that cannot process input from an assistive device may violate the Americans with Disabilities Act.

State-Level Regulation

New York City Local Law 144

Effective July 2023, New York City requires bias audits for automated employment decision tools (AEDTs) used in hiring and promotion. Requirements include:

Bias audit mandate: AEDTs must undergo annual bias audits examining selection rates across protected categories.

Disclosures: Employers must notify candidates that an AEDT will be used and provide the bias audit summary.

Third-party auditing: Bias audits must be conducted by independent third parties.

Documentation: Employers must retain bias audit reports and make them available to the NYC Department of Consumer and Worker Protection upon request.

California Civil Rights Department Guidance

California has issued guidance on AI in hiring, emphasizing:

BFOQ defense: Employers using AI must be able to demonstrate that any screening criteria are bona fide occupational qualifications.

Disclosure obligations: Candidates may have rights to know what data is being collected and how it is being used.

Intersectionality: Analysis must consider compounded disadvantage across multiple protected characteristics.

Illinois AI Video Interview Act

Illinois requires:

Consent: Employers must obtain written consent before using AI analysis of video interviews.

Disclosure: Applicants must be notified about how the AI system evaluates them.

Explanation: Upon request, employers must explain what characteristics the AI analyzes and how they relate to job performance.

Risk Categories

High-Risk AI Applications

Resume screening: Automated systems that filter resumes before human review have high disparate impact risk because they often rely on criteria that correlate with protected characteristics.

Video interview analysis: AI that analyzes facial expressions, tone of voice, or emotional states may disadvantage candidates with disabilities, non-native speakers, or neurodivergent individuals.

Psychometric assessments: Personality tests and cognitive assessments must be validated for specific jobs and may disproportionately screen out protected groups if not carefully designed.

Chatbot screening: Automated initial interview systems may use criteria that disadvantage candidates based on communication style rather than job-relevant qualifications.

Skills testing: AI-proctored tests may disadvantage candidates with disabilities requiring accommodation or those with slower internet connections.

Lower-Risk Applications

Calendar scheduling: Tools that simply find meeting times generally present minimal discrimination risk.

Keyword highlighting: Systems that highlight relevant resume sections for human reviewers present lower risk than automated screening.

Communication routing: Basic routing systems that direct candidate inquiries present minimal civil rights concern.

Employer Liability Framework

Strict Liability Scenarios

Employers may face strict liability for AI-driven discrimination when:

• The AI system explicitly uses protected characteristics in its decision-making
• The employer knew or should have known about discriminatory impact and failed to act
• The employer cannot demonstrate job-relatedness and business necessity

Negligence-Based Liability

Even where discriminatory intent cannot be shown, employers may face negligence claims for:

• Failure to conduct due diligence on AI vendors
• Failure to monitor AI systems for discriminatory outcomes
• Failure to implement bias mitigation measures
• Failure to provide accommodation for disabled applicants

Defenses Available

Business necessity: Demonstrate that the AI system's criteria are job-related and consistent with business necessity.

Job validation: Show that the selection criteria have been validated through proper studies demonstrating predictive validity for job performance.

Less discriminatory alternative: Demonstrate that no less discriminatory alternative was available and equally effective.

Good faith efforts: Document attempts to identify and address bias—may mitigate damages but does not eliminate liability.

Best Practices for Compliance

Pre-Deployment Requirements

Conduct bias audits: Before deploying any AI hiring tool, commission a disparate impact analysis examining selection rates across race, sex, national origin, age, and disability status.

Validate for job-relatedness: Ensure the AI system has been validated as predictive of job performance for the specific position and context.

Review vendor documentation: Obtain and analyze technical documentation explaining how the AI system works, what data it uses, and what validation studies have been conducted.

Assess proxy variables: Ask vendors specifically whether their systems use variables that may correlate with protected characteristics.

Ongoing Monitoring

Track selection rates: Monitor hiring outcomes by demographic group to identify emerging disparate impact.

Audit periodically: Conduct bias audits at least annually and whenever the AI system is updated.

Review system changes: When vendors update AI systems, request documentation of changes and assess whether new bias risks have been introduced.

Monitor complaints: Track candidate complaints about AI systems and investigate potential discrimination patterns.

Contractual Protections

Require indemnification: Negotiate vendor agreements requiring indemnification for claims arising from discriminatory AI outputs.

Secure audit rights: Contract for the right to conduct independent bias audits of AI systems.

Obtain representations: Require vendors to represent that their systems comply with applicable law and do not intentionally discriminate.

Establish remediation obligations: Contract for vendor obligations to remediate identified bias within specified timeframes.

Key Resources

• EEOC Guidance on AI and Automated Systems
• EEOC Uniform Guidelines on Employee Selection Procedures
• NYC Local Law 144 Implementation Guidance
• Illinois AI Video Interview Act
• NIST AI Risk Management Framework

About the Author

Elena Markov is a technology employment attorney specializing in algorithmic discrimination and AI governance. She advises employers on AI hiring compliance and represents individuals in discrimination claims arising from automated employment decisions.

This analysis is for educational purposes and does not constitute legal advice. Employers should consult qualified counsel regarding specific AI hiring tool compliance obligations.