AI screening, accent scoring and race discrimination at work in Great Britain

AI hiring tools work with names, addresses, accents, writing style, qualifications and work history. Several of those can be linked to race as the Equality Act 2010 defines it. That does not make the tools unlawful, but it means an employer using them needs to know what they measure and what they do with the result.

The short answer: race includes colour, nationality and ethnic or national origins (s9(1)). An employer must not treat someone less favourably because of race (s13(1)), and must be able to justify a rule applied to everyone that puts a racial group at a particular disadvantage (s19).

What the characteristic covers

• Race: includes colour, nationality, and ethnic or national origins (s9(1)).
• Racial group: a group of persons defined by reference to race (s9(3)). A racial group can be made up of two or more distinct racial groups (s9(4)).

Direct and indirect discrimination

• Direct (section 13): less favourable treatment because of race cannot be justified (s13(1)-(2)). Less favourable treatment because of race includes segregating a person from others (s13(5)). See direct discrimination.
• Indirect (section 19): a provision, criterion or practice applied to everyone that puts a racial group at a particular disadvantage, and the individual at that disadvantage, unless the employer shows it is a proportionate means of achieving a legitimate aim (s19). See indirect discrimination.
• Harassment (section 26): unwanted conduct related to race that has the purpose or effect set out in section 26 (s26(1)); see harassment related to a protected characteristic.

Where AI tools come in

• Nationality filters: a setting that screens out applicants by nationality applies part of the characteristic directly. Checking the right to work is a separate legal duty and should be applied to everyone in the same way; see right to work checks.
• Names and addresses: scoring that draws on names, postcodes or places of education can be linked to ethnic or national origins.
• Accent and language scoring: speech tools that score pronunciation, fluency or "clarity" in video or phone interviews can disadvantage candidates whose first language is not English or who have a regional or national accent. Where a level of English is genuinely needed for the job, the employer would need to be able to show that the way it is tested is proportionate if it disadvantages a group.
• Face and image tools: if an identity check or video tool performs less well for some skin tones, that can lead to more failed checks or lower scores for those candidates. Ask the vendor what testing has been done.
• Overseas qualifications and experience: a model that gives little or no weight to qualifications or work history from outside the UK applies a criterion that can disadvantage people by national origin.

The employer is responsible for what it does with a vendor's tool. Where a job genuinely requires a particular race, a narrow exception can apply; see occupational requirements.

Practical steps

• Employers: remove nationality and name-based filters, ask vendors how speech and face tools perform across groups, offer a human alternative where a check fails, assess overseas qualifications on their content, and monitor outcomes by group where you lawfully can.
• Candidates and workers: keep the advert, any score and the rejection, note whether the tool assessed your speech or face, and ask what criteria were applied.

For time limits and proof, see time limits and the burden of proof; for the Act's other workplace rules, 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.