AI hiring bias: what UK employers should check

Last reviewed 23 September 2026

General information, not legal advice.

How bias gets into AI hiring tools

• Biased history. A model trained to copy past hiring decisions learns the patterns in them. In 2018 Reuters reported that Amazon scrapped an experimental recruiting tool after it learned to penalise CVs that included the word "women's", as in "women's chess club captain".
• Proxies. A tool that never sees gender or ethnicity can still pick up on things that correlate with them - names, postcodes, schools, career gaps, word choice.
• Language and accent. Tools that judge fluency or "communication" can penalise non-native speakers for roles that don't need fluent English.
• Criteria. Even a perfect tool is unfair if it screens for a requirement the job doesn't need.

What UK law expects

Under the Equality Act 2010, employers are responsible for discrimination in their recruitment - including when a tool does the screening. The main risk is indirect discrimination: a criterion applied to everyone that puts a group sharing a protected characteristic at a particular disadvantage, and that you can't justify as a proportionate means of achieving a legitimate aim. UK GDPR adds requirements for fairness, transparency and safeguards around automated decisions.

In November 2024 the ICO published the results of its audits of AI recruitment tools. It found tools that inferred characteristics such as gender and ethnicity from candidates' names, and some that allowed recruiters to filter candidates by protected characteristics, and made around 300 recommendations to providers. It is worth reading before you buy.

A checklist for any AI screening tool

1. Who decides? The tool should support decisions, not make rejections on its own.
2. What does it use? Only what the applicant submits - or also social media and public profiles?
3. Can you see the reasoning? Ask for a full record of each screen, not just a score.
4. Is it tested for bias? Ask how outcomes are monitored across groups, and to see results.
5. Does it infer characteristics? It shouldn't estimate gender, ethnicity or age, or let anyone filter on them.
6. Language: can applicants screen in their own language, and is English only assessed when the job needs it?
7. Is there a human alternative for applicants who need adjustments?

How BeCareers approaches it

The AI gathers information and summarises it; your team makes the hiring decisions, and any knock-out rules are yours to set. Screening uses only the application, the CV and the conversation - not social media. Every applicant is asked about the same criteria, which you set, in their own language, and every result keeps the full transcript and a record of the model that produced it. We would rather show you how it works than ask you to trust it. See our buyer's checklist.