AI IN HIRING DECISIONS

Introduction

Artificial Intelligence (AI) is increasingly used in recruitment and hiring. Employers may use AI systems to screen CVs, rank applicants, analyse qualifications, conduct video interviews, assess personality or behavioural characteristics, identify suitable candidates, and recommend final hiring decisions. Although AI can improve efficiency and reduce administrative costs, its use creates significant legal issues relating to discrimination, privacy, transparency, accountability, data protection, reasonable accommodation, and procedural fairness.

The central legal principle is that an employer generally cannot avoid employment-law responsibility merely because a decision was made or assisted by an algorithm. If an AI system produces discriminatory hiring outcomes, the employer may still face liability under applicable employment-discrimination laws.

The U.S. Equal Employment Opportunity Commission (EEOC) expressly recognizes that existing employment-discrimination laws apply when AI is used in recruitment, screening and hiring. Protected characteristics include race, colour, religion, sex, national origin, age, disability and genetic information.

1. Meaning of AI in Hiring Decisions

AI in hiring refers to the use of algorithmic or machine-learning systems to assist or make decisions concerning prospective employees.

Common applications include:

Automated CV screening;

Applicant ranking and scoring;

Job-candidate matching;

Automated interview scheduling;

Video-interview analysis;

Personality assessment;

Skills assessment;

Chatbots for recruitment;

Predictive assessment of candidate suitability;

Automated rejection or shortlisting.

For example, an employer may instruct an AI system to evaluate 10,000 applications and select the candidates whose qualifications most closely resemble those of previously successful employees.

The legal problem arises when the historical data itself reflects discrimination or when seemingly neutral criteria operate as a proxy for protected characteristics.

2. AI Bias and Discrimination

AI systems learn patterns from data. If historical hiring data contains discriminatory patterns, an algorithm may reproduce or amplify those patterns.

For example, if an organisation historically hired predominantly men for engineering positions, an AI model trained on those hiring decisions may learn that characteristics associated with male applicants indicate a higher probability of success.

This is commonly referred to as algorithmic bias.

The bias may arise from:

discriminatory training data;

incomplete datasets;

biased variables;

inappropriate proxy variables;

discriminatory historical decisions;

flawed model design;

lack of testing;

inadequate human supervision.

Under employment-discrimination law, a neutral-looking algorithm may still create liability if it disproportionately excludes a protected group without sufficient legal justification.

3. Direct Discrimination

Direct discrimination occurs where an employer intentionally uses a protected characteristic in making a hiring decision.

An AI system could potentially produce direct discrimination if it is deliberately programmed to give different treatment based on:

sex;

race;

religion;

age;

disability;

national origin; or

another legally protected characteristic.

For example, instructing an algorithm to reject women from a particular occupation would not become lawful merely because the instruction was executed by software rather than a human recruiter.

4. Disparate Impact / Indirect Discrimination

A more difficult issue arises where an AI system does not expressly use a protected characteristic but nevertheless disproportionately disadvantages a protected group.

For example, an algorithm may give substantial weight to:

employment gaps;

particular educational institutions;

geographical location;

vocabulary;

working hours;

online behaviour;

previous salary;

particular career histories.

These factors may operate as proxies for protected characteristics.

U.S. employment law recognizes that a facially neutral employment practice may create unlawful disparate impact when it disproportionately affects protected groups and is not sufficiently justified by job-related necessity. The EEOC specifically notes that hiring practices with a negative effect on protected groups require legal scrutiny.

5. Case Law

Case 1: Mobley v. Workday, Inc.

Court: U.S. District Court for the Northern District of California

Importance: One of the most significant emerging cases concerning AI-assisted employment discrimination.

In Mobley v. Workday, Inc., the plaintiff alleged that Workday's algorithmic applicant-screening tools discriminated against applicants on grounds including race, age and disability.

The litigation concerns whether an AI/software provider can face liability where employers use its automated tools in recruitment and hiring.

In 2026, the court allowed the litigation to continue in substantial part, including disparate-impact theories under Title VII, the Americans with Disabilities Act (ADA), and the Age Discrimination in Employment Act (ADEA). The case has also developed to include gender-related and California-law allegations.

Legal principle:
The use of an automated hiring system does not necessarily remove discrimination law from the employment decision. Questions of responsibility may extend to both the employer using the system and, depending upon the facts and applicable law, the technology provider.

Case 2: Griggs v. Duke Power Co., 401 U.S. 424 (1971)

The U.S. Supreme Court established the modern disparate-impact principle.

The employer required applicants for certain jobs to possess a high-school education or pass aptitude tests. Although these requirements appeared neutral, they disproportionately excluded Black applicants and were not sufficiently related to job performance.

The Court held that employment practices that are neutral in form may nevertheless violate Title VII when they disproportionately exclude protected groups and are not justified by business necessity.

Relevance to AI:
An AI hiring algorithm can similarly be challenged where its apparently neutral criteria disproportionately exclude a protected group.

Case 3: Albemarle Paper Co. v. Moody, 422 U.S. 405 (1975)

The U.S. Supreme Court further developed disparate-impact principles and emphasized the importance of validating employment selection procedures.

Relevance to AI hiring:
AI-based assessments should be demonstrably connected to legitimate job requirements. A sophisticated algorithm is not legally justified merely because it is technologically advanced.

Case 4: Washington v. Davis, 426 U.S. 229 (1976)

The U.S. Supreme Court distinguished discriminatory intent from discriminatory effects.

The case established that discriminatory impact alone does not automatically establish a constitutional violation under the Equal Protection Clause.

Relevance to AI:
AI hiring disputes require careful identification of the applicable statutory or constitutional framework. A discriminatory outcome may be legally significant under employment statutes even where intentional discrimination is difficult to prove.

Case 5: McDonnell Douglas Corp. v. Green, 411 U.S. 792 (1973)

The Supreme Court established an important evidentiary framework for employment-discrimination claims.

The framework generally involves:

establishing a prima facie case;

requiring the employer to articulate a legitimate, non-discriminatory reason;

allowing the claimant to demonstrate that the stated reason is pretextual.

Relevance to AI:
Where an employer relies upon an automated hiring recommendation, questions may arise concerning what constitutes the employer's legitimate reason and how an applicant can challenge an algorithmic decision.

Case 6: Ricci v. DeStefano, 557 U.S. 557 (2009)

The Supreme Court considered an employment testing system and the employer's response to racially disproportionate results.

The Court emphasized that employers must carefully navigate the relationship between employment testing, discrimination law and concerns about disparate impact.

Relevance to AI:
AI-based recruitment assessments should be evaluated before deployment rather than treated as automatically lawful because they are statistically sophisticated.

Case 7: EEOC v. Griggs Principle – Application to Employment Testing

The broader body of U.S. employment-testing jurisprudence establishes that selection tests should be related to the requirements of the job and should not unnecessarily exclude protected groups.

The EEOC similarly explains that employers using tests during recruitment must ensure that the test is necessary and job-related and does not unlawfully exclude protected applicants.

Relevance to AI:
An AI personality test, video-interview assessment or automated ranking system may require validation and monitoring where it functions as an employment-selection test.

6. Amazon AI Recruitment Tool – Important Real-World Example

Although not a reported judicial decision, Amazon's abandoned AI recruitment experiment is an important practical example.

Reports in 2018 stated that Amazon developed a machine-learning recruitment tool trained on approximately ten years of previous applicant data. The system reportedly learned patterns from a historically male-dominated applicant pool and penalized certain indicators associated with women, including references to women's organisations and women's colleges.

Amazon ultimately abandoned the recruitment system after concerns arose regarding gender bias.

Legal significance:
The example demonstrates the principle of "garbage in, garbage out": an AI model trained on historically biased employment decisions may reproduce those patterns.

It also demonstrates why removing explicit gender information may not be sufficient. Algorithms can identify indirect proxies for protected characteristics.

7. Disability Discrimination and AI Hiring

AI recruitment can create special problems for persons with disabilities.

For example:

automated video analysis may disadvantage applicants with speech impairments;

facial-analysis technology may perform differently across applicants;

timed digital assessments may disadvantage certain disabilities;

personality tests may produce inappropriate exclusions;

automated communication systems may not accommodate applicants requiring accessibility support.

The EEOC states that employment-discrimination protections continue to apply when AI is used and that reasonable accommodation obligations may still arise during recruitment and hiring.

Therefore, employers should provide appropriate alternative assessment mechanisms where legally required.

8. Transparency and Explainability

Traditional recruitment decisions can be explained by a human recruiter.

AI systems can be more difficult to explain because complex machine-learning models may use thousands of variables.

This creates several legal questions:

Why was the applicant rejected?

What factors affected the score?

Was a protected characteristic used?

Was a proxy used?

Was the algorithm validated?

Who designed the system?

Who reviewed the output?

Can the applicant challenge the decision?

Transparency is therefore an important component of legally responsible AI hiring.

9. Human Oversight

AI should generally be treated as a decision-support mechanism rather than an unquestionable decision-maker.

Human oversight can involve:

reviewing AI recommendations;

investigating unusual rejection patterns;

conducting discrimination audits;

testing the system before deployment;

reviewing complaints;

maintaining records;

allowing human reconsideration of automated decisions.

Human involvement, however, should be meaningful. Simply placing a human at the end of an automated process does not necessarily eliminate legal responsibility.

10. Employer Liability

A major legal question is:

Who is responsible when AI discriminates?

Potentially relevant actors include:

A. Employer

The employer may remain responsible because it ultimately controls the recruitment process.

B. AI Vendor

A technology provider may face contractual, statutory or other legal claims depending on the jurisdiction and facts.

C. Recruitment Agency

Where an agency uses automated screening, responsibility may depend upon the agency's role and applicable employment laws.

D. Human Decision-Maker

A recruiter may face consequences where the recruiter knowingly relies upon discriminatory algorithmic results.

The emerging Mobley v. Workday litigation illustrates precisely this developing question concerning responsibility for algorithmic employment decisions.

11. Data Protection and Privacy

AI hiring frequently requires large amounts of personal information.

Potentially collected information includes:

CV information;

educational history;

employment history;

interview recordings;

biometric information;

personality information;

online information;

behavioural data.

The legality of collecting and processing such information depends upon the jurisdiction's privacy and data-protection framework.

Employers should therefore consider:

lawful collection;

purpose limitation;

data minimisation;

security;

retention;

access rights;

transparency;

consent or another lawful basis where required.

12. Confidentiality of AI Hiring Data

AI recruitment platforms may process sensitive applicant information through third-party vendors.

Employers should therefore establish contractual safeguards concerning:

confidentiality;

cybersecurity;

data retention;

subcontractors;

data deletion;

access controls;

audit rights;

incident reporting.

A hiring algorithm should not become a mechanism for uncontrolled transfer or reuse of applicant information.

13. Validity of Automated Rejection

An automated rejection is not necessarily unlawful merely because it was generated by AI.

The important questions are:

Was the criterion lawful?

Was it relevant to the job?

Was the system accurately functioning?

Did it create discriminatory effects?

Was the system appropriately validated?

Were reasonable accommodations available?

Was the applicant provided legally required procedural protection?

Thus, AI itself is not inherently unlawful; unlawful discrimination or unlawful data processing through AI creates the legal problem.

14. Compliance Measures for Employers

Employers using AI in hiring should adopt an AI employment compliance framework.

Recommended safeguards:

1. Pre-deployment testing
Test the system for discriminatory outcomes before using it.

2. Job-related validation
Ensure that selection criteria correspond to genuine job requirements.

3. Bias audits
Regularly evaluate outcomes across protected groups.

4. Human review
Maintain meaningful human oversight.

5. Accessibility
Provide appropriate accommodations for applicants with disabilities.

6. Documentation
Maintain records concerning the system's design, testing and use.

7. Vendor due diligence
Assess the AI provider's methodology and compliance practices.

8. Transparency
Inform applicants where AI is materially involved in recruitment where required or appropriate.

9. Complaint mechanism
Provide a mechanism through which applicants can challenge potentially erroneous or discriminatory decisions.

10. Continuous monitoring
AI systems should be monitored after deployment because model performance and applicant populations may change.

15. Key Legal Issues

The principal legal issues concerning AI in hiring decisions are:

Conclusion

AI in hiring decisions represents a major development in modern employment law. Automated recruitment can increase efficiency and consistency, but it can also reproduce historical discrimination, create indirect discrimination, disadvantage persons with disabilities, and make recruitment decisions difficult to explain.

The emerging Mobley v. Workday litigation demonstrates that courts are beginning to confront the question of how traditional employment-discrimination law applies to AI-driven recruitment systems.

The established principles from cases such as Griggs v. Duke Power Co., Albemarle Paper Co. v. Moody, McDonnell Douglas Corp. v. Green, and Ricci v. DeStefano remain particularly relevant because they provide legal principles concerning employment testing, disparate impact, discriminatory motivation and selection procedures.

Therefore, the central legal principle is that automation does not eliminate employer responsibility. AI-based hiring should be job-related, non-discriminatory, appropriately validated, accessible, transparent to the extent required by law, and subject to meaningful human oversight.