Artificial intelligence is rapidly reshaping recruitment, from screening applications to supporting candidate selection. Under the EU AI Act, certain AI systems used in employment are classified as high-risk. Yet some of the corresponding regulatory requirements will apply at a later stage. For businesses, this creates a transitional period – but certainly not a licence for uncontrolled automation.

Artificial intelligence is transforming recruitment on both sides of the labour market. Candidates increasingly use generative AI to draft cover letters, refine CVs and tailor applications to specific vacancies. Employers, meanwhile, are deploying algorithms to process large volumes of applications, compare candidates and accelerate selection procedures.

As AI becomes more deeply embedded in recruitment, questions of transparency, accountability and oversight become increasingly important. Once an algorithm influences who progresses through an application process, flawed data, inappropriate evaluation criteria or systemic bias can have direct consequences for an individual’s career opportunities.

Recruitment AI as a High-Risk Application

The EU AI Act reflects the sensitivity of such applications by classifying certain AI systems used in recruitment, candidate selection and employment-related decision-making as high-risk.

The European regulatory framework subjects these systems to enhanced requirements, including risk management, documentation, data governance and effective human oversight. The underlying principle is clear: the greater the potential impact of an automated system on individuals, the stronger the safeguards surrounding its use must be.

However, the relevant obligations do not all become applicable simultaneously. Parts of the high-risk regime take effect later, giving companies additional time to prepare. This should not be mistaken for a regulatory vacuum.

Existing obligations under data protection, employment and anti-discrimination law remain relevant regardless of whether decisions are made entirely by humans or supported by algorithms. Introducing AI into recruitment therefore does not transfer responsibility from the employer to the technology.

When Algorithms Scale Existing Bias

Automation should not be confused with objectivity. AI systems operate on the basis of data, models and evaluation criteria that may themselves reflect historical inequalities or unsuitable assumptions.

If past recruitment decisions are used to train or configure automated systems, existing patterns of disadvantage may be reproduced rather than eliminated. What distinguishes algorithmic bias from individual human bias is its potential scale. A subjective decision by a recruiter may affect a single applicant; a poorly designed automated system can apply the same problematic logic to thousands of candidates.

At the same time, human decision-making is hardly free from bias. Time pressure, expectations, personal experience and unconscious preferences can all influence recruitment decisions.

The central issue is therefore not whether humans or machines are inherently better decision-makers. It is how responsibility, automation and oversight are distributed between them.

Human Oversight Must Be Meaningful

AI can provide substantial benefits to HR departments. It can structure information, identify relevant qualifications and help organisations process large applicant pools more consistently. Yet the greater its influence over candidate selection, the more important meaningful human supervision becomes.

Human oversight cannot simply mean approving an algorithmic recommendation at the end of the process. Recruiters must be able to understand, challenge and, where necessary, override automated outputs.

Ultimate responsibility for employment decisions remains with the organisation.

To exercise that responsibility effectively, companies first need visibility over where AI is being used. Automated functions are increasingly integrated into applicant tracking systems, HR platforms and cloud-based services, sometimes without being immediately perceived as standalone AI applications.

Creating an internal AI inventory can therefore be an essential first step towards effective governance. Organisations need to know which systems they operate, what data those systems process, which decisions they influence and who is responsible for monitoring their performance.

AI Literacy Is Becoming a Compliance Issue

Technical controls alone will not be sufficient. The ability of employees to use and evaluate AI systems appropriately is becoming an important element of corporate governance.

This extends far beyond IT departments. HR professionals and managers need to understand both the capabilities and limitations of the technologies they rely on. They must be able to identify when an automated result requires further scrutiny and when legal, ethical or organisational boundaries may have been reached.

AI literacy is therefore increasingly connected to risk management. An organisation can deploy sophisticated technology and still create substantial exposure if the people using it lack the expertise to recognise unreliable or discriminatory outcomes.

Preparation Instead of Last-Minute Compliance

Companies should not interpret later regulatory deadlines as a reason to postpone AI governance. Technology can be deployed rapidly; robust structures for documentation, accountability and oversight cannot.

Organisations that establish clear rules today – defining which AI systems may be used, what risks they create, how their outputs are reviewed and who remains accountable – will be better positioned for the evolving European regulatory environment.

More importantly, effective governance is not merely about satisfying future compliance requirements. It can improve the quality, consistency and credibility of recruitment decisions while strengthening trust among candidates and employees.

The decisive question is therefore no longer whether artificial intelligence will become part of recruitment, but under what conditions it should be allowed to influence employment decisions. Efficiency cannot come at the expense of transparency, fairness and human accountability. The transitional period before the high-risk framework takes full effect should be used to build the governance structures that responsible AI recruitment will ultimately require.