Using AI In Recruitment: How To Avoid Discrimination Claims

Recruitment is one of the highest-risk areas for AI use in any organisation. It is also one of the areas where AI tools are being adopted most quickly, often with little scrutiny of what they actually do or how they reach their conclusions. The combination of speed, apparent objectivity, and legal exposure makes this a topic that HR professionals and employers cannot afford to approach without a clear understanding of the risks involved.

The central problem is not that AI recruitment tools are inherently unlawful. It is that they are treated as neutral when they are not, and that employers assume the legal responsibility sits elsewhere when it does not.

How AI is being used in recruitment

AI has found its way into almost every stage of the recruitment process. At the top of the funnel, tools scan and rank CVs, filtering large volumes of applications against criteria the employer has defined, or that the system has inferred from previous hiring data. Further along the process, AI-powered platforms conduct initial screening interviews, assessing candidates through video analysis, voice pattern recognition, or responses to automated questions. Some tools claim to assess personality, cultural fit, or future performance potential. Others generate ranked shortlists, flag candidates for progression, or recommend salary offers based on market data and candidate profiles.

Each of these applications carries legal risk. Some carry more than others. But the common thread running through all of them is that the employer remains fully responsible for the outcomes they produce.

Why AI is not neutral

The idea that algorithmic decision-making removes human bias is one of the most persistent and most legally dangerous misconceptions in this area. AI systems do not arrive at decisions independently. They are trained on data, and the patterns they learn reflect the patterns in that data. Where historical hiring decisions have favoured certain candidates over others, whether by gender, age, ethnicity, educational background, or any other characteristic, a model trained on that history will tend to reproduce those patterns. It has learned, in effect, that certain profiles lead to successful hires, because those are the profiles that were hired.

This is not a malfunction. It is the system working as designed. The problem is that the design embeds discrimination without anyone having chosen to discriminate. An employer using such a tool is not insulated from liability by the fact that no human being made a biased decision. Under the Equality Act 2010, what matters is the outcome. If a protected characteristic is a factor in why a candidate was treated less favourably, a claim can follow regardless of whether that factor was applied by a person or a process.

Age discrimination deserves particular mention. Many AI recruitment tools are trained on data that correlates years of experience, graduation dates, or career trajectory with suitability for a role. These proxies can operate as indirect discrimination against both younger and older candidates in ways that are not immediately visible in the tool’s outputs. The same is true of tools that analyse language patterns or communication styles, which can disadvantage candidates whose first language is not English, or whose cultural background shapes how they express themselves.

The legal framework

The Equality Act 2010 prohibits both direct and indirect discrimination across all nine protected characteristics in the context of recruitment. Direct discrimination, treating a candidate less favourably because of a protected characteristic, is relatively straightforward to understand. Indirect discrimination is where AI creates the greater risk. It arises where an employer applies a provision, criterion, or practice that puts people with a particular protected characteristic at a disadvantage compared to others, and where that cannot be objectively justified.

An AI tool that consistently ranks male candidates above female candidates for technical roles, or that filters out applicants over a certain age bracket, may be applying exactly such a provision. The employer’s lack of awareness that the tool was doing so is not a defence. Nor is the argument that the tool was supplied by a third party. The employer is the data controller and the decision-maker. Responsibility does not transfer with the software licence.

The right to an explanation under UK GDPR adds a further layer of obligation. Where AI is used to make or materially inform recruitment decisions, candidates have rights in relation to automated processing that affects them. An employer who cannot explain why a candidate was rejected, beyond the fact that the system ranked them below others, is in a legally uncomfortable position.

Where employers most commonly go wrong

The errors that create legal exposure in AI-assisted recruitment tend to follow recognisable patterns.

The first is deploying tools without interrogating how they work. Procurement decisions are often made on the basis of commercial terms and user experience rather than a serious examination of what the tool is optimising for and what data it was trained on. Employers who cannot answer basic questions about their recruitment AI, including what criteria it applies, how those criteria were derived, and whether the tool has been tested for discriminatory outcomes, are operating without adequate oversight.

The second is treating AI outputs as decisions rather than inputs. A ranked shortlist produced by an AI tool is not a hiring decision. It is a starting point for human assessment. Where recruiters treat AI rankings as effectively final, or where the practical reality is that candidates outside the top tier are never meaningfully considered, the tool has become the decision-maker in all but name. That creates both legal exposure and governance risk.

The third is failing to monitor outcomes. Even a tool that performs acceptably at the point of deployment can drift over time, or can produce discriminatory patterns that only become visible across a larger dataset. Employers who do not periodically review the demographic profile of candidates progressing through AI-assisted stages of their recruitment process are missing the clearest available signal that something may be wrong.

The fourth is inadequate record-keeping. If a discrimination claim arises, the employer will need to demonstrate what criteria were applied, how the tool was used, what human oversight existed, and what decisions were made at each stage. Employers who cannot reconstruct that picture are significantly disadvantaged in any proceedings.

Practical steps for employers

Managing discrimination risk in AI-assisted recruitment does not require abandoning the technology. It requires using it with appropriate oversight and governance in place.

Before deploying any AI recruitment tool, employers should carry out due diligence on the supplier. That means asking what data the tool was trained on, how it has been tested for bias, what protected characteristics its outputs have been audited against, and what the supplier’s position is on liability where discriminatory outcomes occur. A reputable supplier should be able to answer these questions. One that cannot give the employer pause.

Data protection obligations also require attention at this stage. A data protection impact assessment should be completed before deploying any AI tool that processes personal data about candidates. Where automated decision-making is involved, the legal basis for that processing must be identified and candidates must be informed that AI is being used in the process.

Human oversight must be genuinely meaningful throughout. This means ensuring that AI outputs are treated as one input among several, that trained recruiters apply independent judgement to candidate assessment, and that no candidate is effectively excluded from consideration solely on the basis of an algorithmic ranking. Where video interview analysis tools are used, particular care is required: the scientific basis for claims about personality or cultural fit assessment from video footage is contested, and reliance on such assessments without proper scrutiny carries significant legal and reputational risk.

Employers should also establish a monitoring framework that tracks the demographic outcomes of AI-assisted recruitment stages over time. If candidates sharing a protected characteristic are being filtered out at a particular stage at a disproportionate rate, that pattern needs to be investigated and addressed. Waiting for a claim to surface is not a strategy.

Clear records of the recruitment process, including the criteria applied, the AI tools used, the outputs generated, and the human decisions made at each stage, should be maintained and retained. If a candidate brings a discrimination claim, those records will be central to the employer’s response.

A note on candidate transparency

Candidates have a legitimate interest in knowing that AI is being used to assess them. Beyond the legal requirements around automated decision-making under UK GDPR, there are sound practical and reputational reasons for employers to be open about their use of AI in recruitment. Employers who are transparent about how AI is used, what role it plays, and how human oversight operates are better placed both to meet their legal obligations and to maintain confidence among the candidates they are trying to attract.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.