Trigger: AI may affect hiring, screening, or workforce decisions.
HR AI Evidence Passport
Workforce AI must be reviewable
before it decides.
Fairness signals, explainability, and human oversight conditions for AI that screens, ranks, or evaluates candidates and employees.
Sensitive context
Why HR AI is one of the most consequential AI contexts.
AI tools that screen CVs, rank candidates, score interview recordings, or influence promotion decisions have direct consequences for people's livelihoods and opportunities. Unexplained outputs, demographic disparities, and opaque scoring undermine fairness, dignity, and legal obligations.
Data categories in scope
People affected
Risk scenarios
What typically goes wrong.
Specific failure modes seen in this sensitive context — without structured evidence.
A CV screening model ranks candidates with gap years lower without any documented rationale.
No explainability. No fairness audit. Adverse impact on candidates from protected characteristic groups unexamined. Employment tribunal exposure.
An interview analysis AI scores candidates on 'confidence' derived from audio features.
No methodological validation. Protected characteristic proxies possible in voice features. No evidence of human override. GDPR Art. 22 unaddressed.
A workforce reduction model identifies at-risk roles with no demographic disparity analysis.
No Group Disparity Analysis Signal Receipts. Adverse impact on older workers or protected groups undocumented. Post-redundancy legal exposure.
A performance management AI flags employees for performance improvement plans.
No Feature Attribution Evidence. No human review gate. Employees cannot receive a meaningful explanation. GDPR subject access request reveals no interpretable output.
Scope
What needs a Passport.
Stakeholder workflow
From trigger to access decision.
Trigger
AI system in scope
Evidence Request
Passport initiated
Review
DPO · CISO · Specialist
Decision
Access condition set
Monitor
Tide sweeps · Renewal
Trigger
AI system in scope
Evidence Request
Passport initiated
Review
DPO · CISO · Specialist
Decision
Access condition set
Monitor
Tide sweeps · Renewal
CHRO / HR Director
“AI is being used in hiring, evaluation, or workforce decisions.”
Require Passport for every HR AI system. Human oversight gate and fairness signals are mandatory before sign-off.
DPO
“Candidate or employee personal data is processed by an AI model.”
Confirm GDPR legal basis, DPA, and Art. 22 automated decision-making status. Block clearance until confirmed.
Legal / Employment Counsel
“An AI system may influence redundancy, promotion, or disciplinary outcomes.”
Review fairness section of Passport. Document exposure before any outcome communicated to employees.
Access decisions
Context Visa conditions.
The access decisions that apply in this sensitive context — and the evidence conditions that produce them.
- AI influences hiring, promotion, or redundancy decisions
- Human HR professional must review before decision recorded
- Candidate or employee right to explanation documented
- Fairness Signal Receipts absent or outdated
- Explainability not configured for adverse decisions
- Protected characteristic proxy risk not assessed
- Fairness audit complete with subgroup breakdown
- Feature Attribution Evidence available per decision
- Human review gate active for all people decisions
- Annual fairness review required
- Demographic disparity above threshold with no remediation
- No explainability for decisions subject to employee challenge
- Raw protected characteristic data in model features without legal basis
Measurement
Evidence families we can structure.
The measurable evidence categories relevant to this context and the evidence signals they produce.
Fairness & Adverse Impact
Demographic parity, equalised odds, disparate impact ratio across protected characteristics including gender, ethnicity, age, and disability.
Explainability
Feature importance and prediction explanations for individual candidate or employee decisions — supporting employee right to explanation.
Performance & Drift
Model accuracy and data drift evidence — ensuring the model behaves consistently as the applicant pool changes over time.
Privacy
GDPR legal basis for processing application and employee data, DPA status, and confirmation of data minimisation.
Human Oversight
Evidence that human HR professionals review AI outputs before decisions affecting candidates or employees are recorded.
Access Control
Policy-as-code limiting which HR roles can invoke which AI systems and restricting model feature access.
Honest scope
What remains not assessable.
AffectLog does not overclaim. These items require external expertise, regulatory process, or long-term study.
Legal conclusion of unlawful discrimination
Legal determinations of discrimination require employment tribunal or court adjudication — not an evidence platform.
Instead: Refer measured disparities and evidence to employment lawyers for legal interpretation.
Validity of interview AI for hiring quality prediction
Predictive validity of HR AI tools for job performance requires industrial/organisational psychology evaluation and longitudinal study.
Instead: Commission an I/O psychology validation study before using interview AI for high-stakes decisions.
GDPR compliance of the HR process as a whole
Lawfulness of the complete HR process requires HR legal review — AffectLog covers the AI-specific technical evidence layer.
Instead: Engage HR legal counsel or an employment law specialist for the full GDPR compliance review.
Example
Sample Passport for this context.
HiringLens Screening Model
Applicant CV ranking and shortlisting · HR / Talent Acquisition
Access conditions
What we will not overclaim
AffectLog does not claim a system is non-discriminatory or legally compliant with employment law. We show measured disparities, data limitations, explainability evidence, and required human oversight conditions — so HR and legal teams have evidence rather than vendor claims.
Common questions
Questions this context raises.
“Our HR AI vendor says their model is bias-free.”
'Bias-free' is not a verifiable claim without evidence. AffectLog structures the specific measurement: Group Disparity Analysis Signal Receipts, demographic parity ratios, subgroup performance breakdowns, and protected characteristic proxy analysis — per model.
“We use AI only as a first filter — humans make all final decisions.”
First-filter AI still affects who reaches human review. Disparate impact can occur at the filtering stage even if the final decision is human. Fairness Signal Receipts at the filter stage are as important as at the decision stage.
Get started
Make HR AI evidence-based
before it influences the next hire.
Map which AI tools are in your HR workflow, identify which need fairness audits, and establish human oversight conditions before any system affects candidates or employees.
AffectLog provides technical and operational evidence. Not employment law advice, non-discrimination certification, or regulatory compliance.