AI usually arrives in hiring one of two ways. A feature that ranks applicants or suggests matches gets switched on in the recruitment system or job board. Or a recruiter pastes a pile of applications into a chat tool and asks which five are strongest. Either way, a tool is now shaping a decision about a person who never chose it, and the law treats that with more care than almost anything else AI does at work.
None of this is legal advice. Your advisers and your HR lead decide the position. Before a tool screens a single candidate, get answers to the seven checks below.
Why hiring is different
Hiring decisions are about people, which is the clearest case in when a plain workflow beats AI of work where people keep the decision. Discrimination law applies whether a person or a tool made the call. If a tool is trained on past hiring, it can carry the patterns of past hiring forward, including the ones nobody would defend. And the candidate usually has no idea a tool was involved, which makes a wrong outcome very hard to challenge.
What regulators have found and said
The UK Information Commissioner's Office audited providers of AI recruitment tools between 2023 and 2024 and published its findings in November 2024. It made 296 recommendations. Among the issues it found were tools that inferred gender and ethnicity from candidates' names, filters that let recruiters screen out people with protected characteristics, and accuracy that had not been properly tested.
In the UK, the rules on decisions made solely by automated means changed from 5 February 2026, as set out in AI and data protection. Safeguards still apply: telling the person, letting them make their case and giving them a human who can look at the decision again.
In the European Union, AI used for recruitment and selection is classed as high-risk under the AI Act. After the 2026 amendment, the obligations for those systems are due from 2 December 2027.
In the United States, New York City requires bias audits and advance notice to candidates for automated employment decision tools, and other states are writing their own rules. Federal anti-discrimination law applies to decisions made with a tool exactly as it does to any other.
Seven checks before a tool screens candidates
- What does it decide? Ranking a shortlist for a person to review is very different from rejecting candidates automatically. When a tool would reject candidates without a person looking, that use needs the highest bar and usually should not happen at all.
- What does it use? Find out which information it reads, whether it infers anything about protected characteristics, and whether filters that could exclude people on those grounds can be switched off.
- What evidence is there that it is fair? Ask the vendor for its bias and accuracy testing, and test it yourself on past roles where you know the outcome.
- Who reviews, and do they have time to? If a person approves the tool's shortlist in seconds, they are not reviewing it. Give reviewers the authority to overturn it and the time to do so, and record that they did.
- Are candidates told? Your privacy notice should say that AI is used in screening and how a candidate can ask for a person to look at their application.
- Has an impact assessment been done? Using AI on candidates' personal data at any volume is the kind of processing an impact assessment is for.
- What happens to the data? Check the vendor's terms on how long candidate data is kept, whether it is used to improve their product and where it is held.
Applications pasted into chat tools
The less visible version of AI in hiring is a recruiter with a personal chat account and a folder of applications. That is candidates' personal data on terms the business never agreed, and an assessment of people with no record of how it was made.
Where AI helps safely
AI is useful in hiring where it supports people rather than judging candidates. That covers drafting a job listing and checking it for wording that puts people off, scheduling interviews and preparing interview questions from the role description. It can also sum up a candidate's own written answers for the interviewer to read alongside them. In each case a person still makes every decision about who goes forward.
Where this fits
Hiring tools belong on the register with their own line, and decisions about people belong in the policy's review rule, both set in Govern. If your recruitment system is adding AI features, treat each one as a new tool and check it before it is switched on.