AI tools that filter on a protected characteristic: direct discrimination in Great Britain

Some AI hiring and scoring tools can be set up to filter people directly: an age range, a graduation year, a gender, a nationality, a gap for maternity leave. Others are trained or prompted in ways that make a protected characteristic part of the decision. In Great Britain, that is the territory of direct discrimination, and for most characteristics the employer cannot justify it.

The short answer: under section 13 of the Equality Act 2010, an employer discriminates if, because of a protected characteristic, it treats someone less favourably than it treats or would treat others (s13(1)). Only for age can the employer show that the treatment is a proportionate means of achieving a legitimate aim (s13(2)).

The test

• Less favourable treatment: rejecting, ranking lower, offering worse terms or not passing someone on, compared with how others are or would be treated (s13(1)). The comparison can be with a real person or with how a hypothetical person would be treated ("treats or would treat").
• Because of a protected characteristic: the characteristic is the reason for the treatment.
• Like with like: there must be no material difference between the circumstances of the two cases (s23(1)). Where the characteristic is disability, those circumstances include a person's abilities (s23(2)(a)).
• The decision maker's own characteristics do not matter: it does not matter whether the person who discriminated has the characteristic (s24(1)).

Special rules in section 13

• Age: treatment because of age is not discrimination if the employer can show it is a proportionate means of achieving a legitimate aim (s13(2)). Specific exceptions for pay based on service, youth rates and enhanced redundancy pay are on age exceptions at work.
• Disability: a person who is not disabled is not discriminated against only because disabled people are treated more favourably (s13(3)). So, for example, a guaranteed interview or an adjusted test for disabled candidates is not direct discrimination against others on that ground alone.
• Marriage and civil partnership: at work, this applies only if the treatment is because it is the person themselves who is married or a civil partner; see marriage and civil partnership at work (s13(4)).
• Race: less favourable treatment includes segregating the person from others (s13(5)).
• Sex: less favourable treatment of a woman includes treating her less favourably because she is breast-feeding; where the person is a man, special treatment given to a woman in connection with pregnancy, childbirth or maternity is not taken into account (s13(6)). For pregnancy and maternity at work, see pregnancy and maternity.
• Gender reassignment absence: section 16 adds a specific rule for absence because of gender reassignment, which must not be treated less favourably than sickness absence.

Combined characteristics

Section 14 of the Act would cover less favourable treatment because of a combination of two protected characteristics, but legislation.gov.uk shows it as prospective, meaning not in force (s14). A claim about, say, a tool that disadvantages older women is therefore brought under the individual characteristics.

Where AI tools come in

• Explicit filters: settings that screen out candidates by age, date of birth, graduation year, sex, nationality or religion apply a characteristic directly. For age, the employer would need to justify it; for the others, there is no general justification defence under section 13.
• Sex: filters and models that favour one sex are covered on sex discrimination at work.
• Age: date of birth and graduation-year filters, experience bands and retirement flags are covered on age discrimination at work.
• Sexual orientation: social media checks, inferred profiles and partner rules in HR systems are covered on sexual orientation at work.
• Race: nationality filters and other settings that screen on colour, nationality or ethnic or national origins are covered on race at work.
• Religion or belief: filters and keyword lists that screen on religion, or on words chosen because they signal it, are covered on religion or belief.
• Prompts and instructions: telling a tool to prefer "young and energetic" candidates, "native" speakers or a particular profile can build a characteristic into the instruction.
• Linked inputs: inputs that do not name a characteristic but are linked to one, such as career gaps or postcodes, can raise indirect discrimination.
• Occupational requirements: where a job genuinely requires a characteristic, a narrow exception can apply; see occupational requirements.

As with other claims, the employer is responsible for what it does with a vendor's tool. For time limits and who has to prove what, see time limits and the burden of proof.

Practical steps

• Employers: check the filters, prompts and fields each tool uses, remove any that apply a protected characteristic unless an exception plainly applies, and record why each remaining criterion is needed.
• Candidates and workers: keep the advert, any screening questions and the rejection, and note any question about age, family, nationality or similar that the tool asked.

For the Act's other workplace rules, see AI and the Equality Act 2010.

This page covers Great Britain and is general information on the law as at 2 November 2026, not legal advice.