Using AI in Recruitment Without Breaking UK Discrimination Law

Where AI has crept into hiring

AI now sits inside most stages of recruitment: CV screening and ranking, chatbot pre-screens, one-way video interviews scored by software, gamified assessments and automated ad targeting. Much of it arrives quietly, because applicant tracking systems keep adding AI features to existing subscriptions. Plenty of UK employers are running algorithmic screening today without ever having made a deliberate decision to adopt it.

The legal position is blunt. Under the Equality Act 2010, the employer is liable for discriminatory outcomes in its hiring process, not the software vendor. 'The algorithm decided' is not a defence, and compensation for discrimination at an employment tribunal is uncapped. That combination is why this deserves more care than the average software purchase.

How the Equality Act applies to an algorithm

The Act protects nine characteristics, including age, disability, race, sex, pregnancy and maternity, and religion or belief. Direct discrimination by an AI tool is rare; the real danger is indirect discrimination, where a seemingly neutral criterion disadvantages a protected group. Algorithms are exceptionally good at finding proxies for characteristics they were never shown.

• Penalising career gaps disadvantages women returning from maternity leave and people with long-term health conditions
• Scoring speech fluency, pace or accent in video interviews can disadvantage candidates by race or disability
• Timed gamified tests can disadvantage older candidates and disabled candidates
• Weighting postcode, school or previous employer can act as a proxy for race or socio-economic background

Indirect discrimination can be lawful if the practice is a proportionate means of achieving a legitimate aim, but that defence is hard to run if you never tested the tool and cannot explain how it scores people. Ignorance of the mechanism weakens your position; it does not excuse it.

What the ICO expects from you

Recruitment AI processes personal data at scale to make decisions with significant effects, which puts it squarely in high-risk territory under UK GDPR. That means a data protection impact assessment before deployment, not after. The ICO has audited AI recruitment providers and published detailed recommendations, and its message to employers is consistent: you must understand the tool, tell candidates it is being used, and be able to explain its outputs.

• Complete a DPIA covering the tool, its data flows and its bias risks before go-live
• Tell candidates clearly, in the job ad or application journey, that AI is used and for what
• Ensure any rejection influenced by AI involves meaningful human review, since Article 22 of UK GDPR gives candidates rights around solely automated decisions with significant effects
• Collect only the data the tool needs, and set retention periods for unsuccessful applicants

Meaningful human review means a person with authority to overrule the machine, time to look at the underlying application, and a record of doing so. A recruiter clicking accept on a ranked list at speed is rubber-stamping, and regulators and tribunals can tell the difference.

Disabled candidates and reasonable adjustments

The duty to make reasonable adjustments applies to the recruitment process itself, not just the job. AI-scored video interviews can disadvantage candidates with speech differences or facial paralysis; timed cognitive games can disadvantage dyslexic and autistic applicants; voice-based chatbots can exclude deaf candidates. If your process relies on any of these, you need a genuine alternative route, a phone call, a standard interview, extra time, and it must be easy to request.

Advertise the adjustment route prominently rather than burying it. A hidden or grudging adjustments process is itself evidence of failure to comply, and it filters out strong candidates you wanted to reach.

The pre-adoption audit checklist

Run this before any AI hiring tool touches a live vacancy:

• 1. Ask the vendor for bias-testing evidence, including results across groups relevant to a UK workforce, and get their claims in writing
• 2. Ask what data the model was trained on and whether it reflects UK hiring or was built for another market
• 3. Run a shadow trial: score one full hiring cycle with the AI alongside your normal human process and compare the two shortlists by sex, age band, ethnicity and disability where you hold that data
• 4. Complete your DPIA and have someone senior sign it off
• 5. Write the candidate-facing transparency wording and the adjustments route before launch
• 6. Define the human review step: who, with what authority, recorded where
• 7. Set retention periods for scores, recordings and rejected applications
• 8. Put audit rights and cooperation duties into the vendor contract
• 9. Name one accountable owner for the tool inside your business
• 10. Diarise a re-test after every significant vendor update

Staying lawful after go-live

Bias is not a one-off test; models drift, vendors ship updates and your applicant pool changes. Review outcomes by protected group at least quarterly, keep the records, and act on anything skewed, since tribunals can order disclosure of exactly this material and an unexplained pattern is worse than a documented, corrected one. Train the recruiters who sit above the tool so they understand what it can get wrong, and keep an eye on ICO guidance, which continues to evolve alongside wider UK proposals on AI regulation.

Used carefully, AI genuinely can reduce the drudgery of high-volume hiring without adding legal risk, but the setup work is where the protection lives. If you want a second pair of eyes on a tool you are evaluating, our team can help.