Explain when the EU AI Act, Local Law 144 and equivalent rules classify a recruitment tool as high risk or subject to audit.
Human Resources & Personal Development
Bias Audits and Human Oversight of AI Tools Used in Recruitment
Equips HR and TA leaders to commission independent bias audits of recruitment AI, interpret adverse-impact results and design human oversight that catches errors before candidates are screened out unfairly.
🗓️Duration5 training days
📚Content4 modules · 8 sessions
🎓On completionAccredited attendance certificate
About the programme
Course Overview
Recruitment teams now rely on AI to screen CVs, rank candidates and even conduct first-round video interviews, but regulators have made clear that automation does not remove legal responsibility for discrimination. The EU AI Act classifies most recruitment and candidate-evaluation systems as high risk, New York City's Local Law 144 requires an independent bias audit before an automated employment decision tool can be used, and equality regulators in several jurisdictions now expect employers to show that a human reviewer, not the algorithm, makes the final call. This course gives HR, talent acquisition and people-analytics professionals a working method for commissioning and reading a bias audit, spotting proxy variables that reintroduce discrimination through the back door, and building oversight that is meaningful rather than a rubber stamp on a score the reviewer never questions. Participants work through adverse-impact calculations, vendor due-diligence questionnaires and override logs using realistic recruitment data, and leave with an audit commissioning brief, a human-oversight protocol and a candidate-notice template they can adapt to their own AI-enabled hiring process.
Expected Learning Outcomes
Commission an independent bias audit and specify the adverse-impact statistics a vendor or auditor must report.
Interpret adverse-impact ratios and selection-rate data across protected groups to judge whether a tool is discriminating.
Identify proxy variables, such as postcode or employment gaps, that let AI models reproduce bias without using protected characteristics directly.
Design a human-oversight protocol that defines when a reviewer must examine, question or override an AI recommendation.
Draft vendor due-diligence questions that test an AI supplier's testing methodology, data sources and update cadence.
Build a candidate notice and override log that support transparency obligations and internal accountability.
Who Should Attend
HR and talent acquisition leaders deploying or renewing AI-enabled screening and selection tools.
People analytics and HR technology specialists who evaluate or configure recruitment algorithms.
Diversity, equity and inclusion practitioners assessing fairness risk in hiring processes.
HR compliance, legal and risk teams responsible for algorithmic accountability.
Procurement specialists negotiating contracts with AI recruitment software vendors.
Recruitment agency and RPO managers deploying AI tools on behalf of client organisations.
Course Modules
Select any module to see its sessions and points.
01Regulation and Risk Behind AI Hiring Tools
2 sessions · 8 points
Session 1Where the Law Now Stands on Automated Hiring
- Compare how the EU AI Act, New York City's Local Law 144 and equality legislation in other jurisdictions each define and regulate automated employment decision tools.
- Identify which recruitment activities, such as CV screening, video-interview scoring and chatbot pre-screening, typically fall inside these definitions.
- Distinguish the obligations that fall on the employer that deploys a tool from those that fall on the vendor that builds it.
- Assess the practical consequences of non-compliance, including audit findings, published disclosure and complaints to equality regulators.
Session 2How Bias Enters Recruitment Algorithms
- Trace how historical hiring data can encode past discrimination into a model trained to predict who succeeds in a role.
- Identify proxy variables, including postcode, university attended and employment gaps, that correlate with protected characteristics.
- Examine publicly reported patterns of resume-screening and video-interview tools producing skewed outcomes by gender, ethnicity or disability across different employers.
- Differentiate bias introduced by training data from bias introduced by feature selection or an unrepresentative validation sample.
02Commissioning and Reading a Bias Audit
2 sessions · 8 points
Session 1Designing the Audit Brief
- Define the scope of an independent bias audit, including which tools, decision points and candidate populations must be covered.
- Specify the demographic categories and sample-size requirements an auditor needs to produce statistically meaningful results.
- Agree an audit cadence and re-audit triggers, such as a model update, a new candidate market or a material change in applicant volume.
- Select an auditor with genuine independence from the vendor to avoid a conflict of interest that undermines the findings.
Session 2Interpreting Adverse-Impact Results
- Calculate and interpret adverse-impact ratios and selection rates across gender, ethnicity, age and disability groups.
- Apply the four-fifths guideline as a screening trigger for further investigation rather than a definitive legal test on its own.
- Distinguish statistically significant disparities from small-sample noise before deciding whether a tool needs remediation.
- Translate audit findings into a remediation plan with an owner, a deadline and a re-test commitment.
03Building Meaningful Human Oversight
2 sessions · 8 points
Session 1Designing the Human-in-the-Loop Protocol
- Define the decision points where a human reviewer must examine an AI recommendation before it affects a candidate.
- Distinguish meaningful review, where the reviewer can see the reasoning and override it, from a rubber-stamp step that only confirms the score.
- Set confidence thresholds that trigger mandatory human review for borderline or low-confidence AI recommendations.
- Train reviewers to recognise when they are anchoring on the AI's recommendation instead of exercising independent judgement.
Session 2Governance, Documentation and Escalation
- Maintain an override log that records every case where a human reviewer changed an AI-generated outcome and why.
- Build an algorithmic risk register that tracks each AI tool in use, its risk classification and its last audit date.
- Draft a candidate notice that explains, in plain language, how automated tools are used and how to request human review.
- Establish an escalation path for candidate complaints about automated decisions that reaches someone with authority to investigate.
04Vendor Management and Continuous Monitoring
2 sessions · 8 points
Session 1Due Diligence Before You Buy or Renew
- Draft due-diligence questions that test a vendor's training data sources, testing methodology and model update frequency.
- Negotiate contract clauses that require vendor cooperation with independent audits and disclosure of material model changes.
- Review vendor-supplied model documentation, such as model cards or technical summaries, for gaps that limit accountability.
- Compare competing vendors on transparency and audit support rather than accuracy claims alone.
Session 2Monitoring Performance After Go-Live
- Set up ongoing monitoring dashboards that track selection rates by demographic group after a tool goes into production.
- Schedule periodic re-audits aligned to model updates, new markets or significant shifts in applicant demographics.
- Brief senior leadership on residual algorithmic risk and the evidence base supporting continued use of a tool.
- Build a decommissioning plan for any AI tool that fails a re-audit or cannot be remediated within an acceptable timeframe.
What the participant receives
4 course modules
A structured syllabus
8 training sessions
across 5 days
32 detailed points
Applied, detailed content
Accredited attendance certificate
On completing the programme
Complete your registration
We will contact you within one business day to confirm.
Ready to start?
Reserve your seat and start building the skill.