"Anti-bias" Hiring Tools: What a Review of Vendor Claims Reveals

Many recruitment tools today advertise bias reduction among their benefits. An academic study examined what these claims actually cover, by looking at what vendors disclose — and what they don't.

The Raghavan, Barocas, Kleinberg and Levy (2020) study

Manish Raghavan, Solon Barocas, Jon Kleinberg and Karen Levy published a study in 2020, in the proceedings of the ACM conference on fairness, accountability and transparency, examining eighteen vendors of algorithmic hiring assessment tools (Raghavan, Barocas, Kleinberg & Levy, 2020, *Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency*, 469-481, ACM). Their approach: document what these vendors publicly disclose about their design, validation and bias-detection methods, then examine these disclosures from both a technical and a legal angle.

A gap between the marketing claim and verifiability

The authors' findings are nuanced: bias reduction is frequently used as a selling point, but the precise methods that would make this reduction verifiable are, according to their analysis, rarely documented in a checkable way in the public communications of the vendors studied. The authors also discuss how certain algorithmic "de-biasing" techniques interact, sometimes in complex ways, with anti-discrimination law — a technical choice made to correct one bias can, depending on context, introduce another.

What this means for a buyer

This work does not conclude that algorithmic tools necessarily worsen existing biases — it documents a gap between the marketing claim and the verifiability of the methods behind it. For a buyer, the relevant question isn't therefore "does this tool reduce bias" as a general promise, but "what precise methods are documented, and what data were they validated on". A reasonable requirement, based on this work, is to ask any AI-assisted recruitment vendor for **verifiable documentation** of its bias detection and correction methods — rather than relying on the marketing claim alone. This is a methodological requirement that connects directly to the data governance and technical documentation obligations set out in the European regulatory framework for high-risk AI systems in recruitment.

Clarentia's position

At Clarentia, we believe that verifiable documentation of the criteria and methods used should be available for the recruiter to consult, rather than summarized as a general marketing claim about bias reduction.

Source cited

Raghavan, M., Barocas, S., Kleinberg, J., & Levy, K. (2020). Mitigating bias in algorithmic hiring: Evaluating claims and practices. *Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency*, 469-481. ACM.