Counterfactual, Per-Decision Bias Auditing for Automated Hiring: Localizing and Explaining Disparate Impact in Applicant Tracking Systems

Automated applicant tracking systems (ATS) are widely used in hiring, and recent litigation and regulatory pressures necessitate robust auditing for bias. Current bias assessment methods are limited; group fairness metrics lack individual decision granularity, and local explainers do not align with legal bias standards. A novel method, the AI Bias Firewall (AIBF), is proposed to address these gaps by auditing individual hiring decisions through counterfactual analysis to identify and explain disparate impact.

Why it matters

The increasing reliance on automated systems for critical functions like hiring introduces significant legal and reputational risks if bias is not effectively managed and auditable. Developing robust, legally-aligned methods for bias detection is crucial for maintaining public trust, ensuring fair employment practices, and mitigating potential litigation or regulatory penalties for organizations utilizing such technologies.

Key insights

• Automated applicant tracking systems are increasingly central to candidate progression in hiring processes.
• Legal and regulatory frameworks now require auditable decisions from these automated systems, particularly concerning bias.
• Existing bias auditing tools are insufficient, either providing population-level summaries without individual causality (group fairness metrics) or single-prediction explanations without alignment to legal bias standards (local explainers).
• The AI Bias Firewall (AIBF) offers a per-decision bias auditing method by neutralizing protected-attribute proxies and measuring counterfactual shifts in scoring.
• AIBF aims to localize and explain disparate impact, bridging the gap between broad statistical measures and individual prediction attribution in bias detection.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.21537