Bias in AI: How Discrimination Laws Apply to Algorithms

Published on: | by

Artificial Intelligence (AI) is transforming industries from finance to hiring and healthcare to law enforcement. Algorithms now help decide who gets loans, jobs, parole, and medical treatment. But while AI promises efficiency and objectivity, it can also replicate or even amplify existing biases hidden in the data it’s trained on to utilize.

Bias in AI isn’t just a technical flaw since it’s a legal risk. Across the United States and globally, discrimination laws that were written long before machine learning now apply to AI-driven decision-making. Businesses deploying AI must ensure compliance with statutes like:

  • Title VII of the Civil Rights Act (employment discrimination)
  • Americans with Disabilities Act (ADA)
  • Fair Housing Act (FHA)
  • Equal Credit Opportunity Act (ECOA)
  • California Fair Employment and Housing Act (FEHA)
  • State biometric and privacy laws

This article explores how discrimination laws apply to algorithms, real-world examples, and compliance best practices with a focus on AI bias, algorithmic discrimination, anti-discrimination laws, and ethical AI compliance.

What is AI Bias?

AI bias occurs when an algorithm produces results that are systematically prejudiced due to erroneous assumptions, flawed data, or the way the AI model is built. Common types include:

  1. Historical Bias: Training data reflects historical inequalities (e.g., past hiring patterns favoring men).
  2. Sampling Bias: Training data isn’t representative of the full population.
  3. Label Bias: Human-labeled data includes subjective or prejudiced decisions.
  4. Proxy Bias: An algorithm uses a variable that indirectly correlates with a protected characteristic (e.g., ZIP code as a proxy for race).

As such, even without malicious intent, AI systems can create disparate impacts — a legal trigger under many discrimination laws.

How Discrimination Laws Apply to AI

Most anti-discrimination statutes do not explicitly mention AI or algorithms, but courts and regulators treat algorithmic decision-making the same as human decision-making when determining liability.

  1. Title VII of the Civil Rights Act of 1964
  • Scope: Prohibits employment discrimination based on race, color, religion, sex, or national origin.
  • Relevance to AI: If an AI-powered hiring tool screens out applicants disproportionately based on a protected trait — even unintentionally — an employer may be liable for disparate impact discrimination.
  • Example: If an algorithm favors candidates with certain college degrees that historically have fewer minority graduates, the system may violate Title VII unless the employer can prove business necessity.
  1. Americans with Disabilities Act (ADA)
  • Scope: Prohibits discrimination against individuals with disabilities in employment, public accommodations, and more.
  • Relevance to AI: AI systems used for recruiting or performance evaluations that fail to accommodate disabilities — for example, video interview AI that penalizes lack of eye contact — may violate the ADA.
  1. Equal Credit Opportunity Act (ECOA) & Fair Housing Act (FHA)
  • Scope: ECOA prohibits discrimination in lending; FHA prohibits discrimination in housing.
  • Relevance to AI: Credit scoring and mortgage approval algorithms must avoid disparate impacts based on race, gender, marital status, age, or other protected characteristics.
  • Example: In 2022, the Consumer Financial Protection Bureau (CFPB) warned lenders that they must provide specific, understandable reasons for algorithmic denials under ECOA’s “adverse action” requirements.
  1. State-Level Anti-Discrimination Laws
  • California Fair Employment and Housing Act (FEHA)
  • New York City Local Law 144 (bias audits for automated hiring tools)
  • Illinois Artificial Intelligence Video Interview Act

Note: These laws add specific compliance requirements for businesses using AI in hiring and decision-making.

Regulatory and Enforcement Trends

EEOC Guidance (2022 & 2023)

The U.S. Equal Employment Opportunity Commission issued technical guidance warning employers that AI tools must comply with Title VII and the ADA. The EEOC is actively investigating AI hiring discrimination claims.

FTC Enforcement

The Federal Trade Commission has stated it will use its Section 5 authority to address unfair or deceptive practices involving biased AI. Misrepresenting AI fairness or failing to mitigate known bias can result in enforcement action.

EU AI Act

While not U.S. law, the EU AI Act imposes risk-based requirements for AI systems, including transparency, data governance, and bias mitigation. U.S. companies serving EU residents must comply.

Real-World Cases and Investigations

  • Amazon’s AI Recruiting Tool (2018): Discarded after it was found to downgrade resumes containing the word “women’s.”
  • HireVue Facial Analysis: Faced criticism and regulatory inquiries over possible bias against people with disabilities or certain ethnic backgrounds.
  • Apple Card Gender Bias Allegations (2019): New York regulators investigated claims that women received lower credit limits than men with similar profiles.

Note: These examples illustrate that AI bias has tangible legal and reputational consequences.

Legal Theories of Liability for AI Bias

  1. Disparate Treatment: Intentional discrimination by designing or training an algorithm to favor or disfavor certain groups.
  2. Disparate Impact: Neutral algorithmic criteria that disproportionately affect a protected group, without a strong business necessity justification.
  3. Failure to Accommodate: AI systems that fail to make reasonable adjustments for individuals with disabilities.

Best Practices for AI Compliance and Bias Mitigation

Businesses can reduce legal exposure by following bias audit and documentation practices.

  1. Conduct Pre-Deployment Bias Audits

Test AI models for disparate impacts before rollout. Include independent third-party reviews for high-stake use cases (hiring, lending, housing).

  1. Use Representative Training Data

Ensure datasets reflect the diversity of the population affected by the AI’s decisions. Avoid over-reliance on historical data that encodes past discrimination.

  1. Maintain Transparency

Document the AI system’s purpose, data sources, variables, and testing results. Some laws (like NYC Local Law 144) require public bias audit summaries.

  1. Enable Human Oversight

Avoid “black box” AI in high-impact decisions. Provide a process for human review of contested outcomes.

  1. Provide Notice and Explanation

When AI is used to make a decision, disclose its use and give clear and specific reasons for adverse decisions (required under ECOA and some state laws).

Turning Legal Compliance into a Competitive Advantage

Bias in AI is both a compliance risk and a business risk. Discrimination laws — from Title VII to ECOA to state bias audit requirements — apply to algorithms just as they do to human decisions. Organizations that proactively identify, mitigate, and document AI bias not only reduce legal exposure but also build public trust, improve decision quality, and strengthen brand reputation. In 2025 and beyond, AI compliance is no longer optional since it’s a core part of responsible innovation. Businesses that embed fairness, transparency, and legal awareness into their artificial intelligence strategies will lead in ethics and market trust.

Finally, if your organization uses AI in hiring, lending, housing, or other high-stakes areas, our legal team can help you conduct AI bias audits, draft AI use policies and compliance programs, respond to regulatory inquiries, and train your team on AI ethics and legal obligations. Please contact our law firm to speak with an artificial intelligence attorney.