The Promise and Risk of AI in Recruitment
Automating candidate selection processes offers enterprises real benefits: shorter recruitment time, reduced administrative burden on recruiters, and the ability to process more applications with the same resources. However, the history of AI deployments in HR is full of cautionary tales — systems trained on historical data replicate existing biases, penalizing candidates for reasons unrelated to competence. An algorithm trained on data from an organization that historically hired mainly people from specific universities or demographic groups will learn to discriminate identically to previous human decisions — but faster and at massive scale.
EU AI Act and High-Risk Systems
The EU Artificial Intelligence Act explicitly classifies AI systems used in recruitment, candidate selection, and employee management as high-risk systems. This means obligations significantly exceeding standard software deployment: a mandatory risk management system, technical documentation, registration in the EU database, ensuring human oversight of every decision, and the ability to explain the basis of a decision to a candidate who requests it. Organizations unaware of these requirements risk financial and reputational sanctions.
- Mandatory discrimination risk assessment before deployment and during use
- Training data documentation — where it comes from, how representative it is, what biases it may encode
- Explainability mechanisms — candidates must be able to learn why the system made a specific decision
- Human oversight — automatic application rejections without human review are impermissible
- Regular audits for discriminatory patterns in results
Responsible AI Applications in HR
Contrary to appearances, the EU AI Act does not prohibit using AI in recruitment — it sets standards for responsible use. There are a number of applications that deliver value with low discrimination risk. Automated formal processing — verification of documentation completeness, checking whether minimum formal requirements specified by the organization are met — reduces administrative burden without interfering with the substantive evaluation of candidates. Planning support — analysis of historical data regarding recruitment channel effectiveness, time-to-hire, and sources of top performers — provides valuable strategic information.
Bias Mitigation — Specific Techniques
Organizations that choose to deploy AI supporting substantive selection should apply proven bias mitigation techniques. Input data anonymization — removing name, gender, age, photo, and university information before processing — eliminates the most common discrimination vectors. Diverse training data and regular tests for demographic parity in results help detect system drift toward discriminatory patterns. Every system decision should be presented to the recruiter with a justification referencing specific, measurable competencies, not uninterpretable algorithm scores.
ESKOM.AI approaches AI applications in the HR domain with full regulatory awareness stemming from the EU AI Act. Systems requiring high-risk classification are deployed with the complete set of required documentation, oversight mechanisms, and audit procedures, ensuring clients not only the benefits of automation but also confidence in regulatory compliance.