AI and discrimination at work under US federal law

AI tools are no longer only screening CVs. Employers use them to suggest who gets promoted, flag who should be disciplined, rank people for layoffs or set shift and pay patterns. In the US, the main federal anti-discrimination laws were written long before these tools, but their words cover any practice that affects pay, terms or conditions of employment, whoever or whatever produces it. This page sets out what those statutes say, taken from the US Code, and what changed in federal enforcement in 2025. For hiring and the state laws that name AI directly (New York City, Illinois, Colorado, California), see AI in hiring.

For New York City's bias audit law in detail, see NYC Local Law 144 and bias audits.

The short answer: Title VII, the Americans with Disabilities Act and the Age Discrimination in Employment Act apply to decisions on pay, promotion, discipline and discharge, not only hiring, and nothing in them exempts decisions made with software. Title VII still contains its disparate impact rule in the statute. In April 2025 an executive order told federal agencies to move away from disparate impact enforcement, which changes what federal agencies pursue but does not amend the statutes.

Title VII: treatment and impact

Section 703(a) of the Civil Rights Act of 1964 makes it unlawful for an employer to discharge anyone, or otherwise discriminate with respect to compensation, terms, conditions or privileges of employment, because of race, color, religion, sex or national origin. It is also unlawful to limit, segregate or classify employees in any way which would deprive or tend to deprive them of employment opportunities or otherwise adversely affect their status because of those characteristics (42 U.S.C. 2000e-2). A system that sorts, scores or ranks staff is a way of classifying them.

Section 703(k) sets out disparate impact. A practice is unlawful if a complaining party shows it causes a disparate impact on one of those grounds and the employer fails to show it is job related for the position and consistent with business necessity, or if the complaining party shows an alternative employment practice and the employer refuses to adopt it. Each challenged practice normally has to be shown to cause the impact separately, but if the elements of a decision process cannot be separated for analysis, the process may be analysed as one practice. That last point matters for AI tools whose inputs are hard to pull apart.

Disability: accommodation and screening out

The ADA bars discrimination against a qualified individual with a disability in advancement, discharge, compensation, job training and other terms and conditions of employment (42 U.S.C. 12112). Two parts of section 12112(b) are directly relevant to AI tools:

• Reasonable accommodation: not making reasonable accommodations to the known limitations of a qualified applicant or employee is discrimination unless the employer shows undue hardship. A worker who cannot use a monitored interface, or whose pace is scored by software, may need an adjustment or a different assessment route.
• Screening out: using qualification standards, employment tests or other selection criteria that screen out or tend to screen out people with disabilities is discrimination unless the criteria are shown to be job related for the position and consistent with business necessity.

Age

The ADEA makes it unlawful to discharge, or otherwise discriminate with respect to compensation, terms, conditions or privileges of employment, because of age, and to limit, segregate or classify employees in ways that deprive them of opportunities or adversely affect their status because of age (29 U.S.C. 623). It is not unlawful where the differentiation is based on reasonable factors other than age (section 623(f)(1)). Tools that weigh recent training, years since a qualification or salary history in layoff or promotion rankings are worth checking against this.

What changed in federal enforcement

Executive Order 14281 of 23 April 2025, published in the Federal Register on 28 April 2025, states a policy to eliminate the use of disparate-impact liability in all contexts to the maximum degree possible. It directed the Attorney General and the Chair of the Equal Employment Opportunity Commission to assess pending investigations, suits and positions under federal civil rights laws, including Title VII, that rely on disparate impact, and to take action consistent with that policy. It also asked the Attorney General to consider whether federal law preempts state laws that impose disparate impact liability (Executive Order 14281, 90 FR 17537).

An executive order directs the executive branch. It does not repeal section 703(k), the ADA's screening out rule or state law, so workers and employers should not read it as removing those rules. Separately, the EEOC technical assistance on AI and employment selection that was published on eeoc.gov is no longer available there (checked 31 October 2026). State laws and agencies remain a separate route, see AI in hiring. For Great Britain, see AI at work and the Equality Act 2010.

Practical steps

• Employers: know which decisions on pay, promotion, discipline and dismissal use AI output. Check results by sex, race, age and disability where you lawfully can, keep the evidence that the criteria are job related, and look for less discriminatory alternatives before you rely on a tool.
• Employers: give a clear way to ask for an accommodation or an alternative assessment, and train managers not to treat a score as the decision. See AI in performance reviews.
• Workers: ask whether software was used in a decision about you, what it measured and who reviewed it. Keep a note of dates and what you were told. Charges of discrimination under these statutes are filed with the EEOC within time limits, so do not wait to seek advice.

See also employee monitoring laws in the US and raising concerns about workplace AI. This page is general information, not legal advice.