AI in HR and recruitment: useful, and legally fraught
Screening, scheduling and drafting — where AI helps hiring teams, and where regulation is closing in fast.
Hiring is a domain where AI can save enormous time and where an unexamined model can cause real harm and real liability. The two are not in tension if you are deliberate about which tasks you automate.
Automate the logistics, not the judgement
Scheduling, chasing, structuring notes, drafting descriptions, summarising an application against stated criteria — this is where most of the wasted time sits, and none of it involves ranking humans.
Ranking candidates is the regulated part
Several jurisdictions now require bias auditing, candidate notification, or both, for automated employment decision tools. Before building any ranking system, establish what applies in every market you hire in — this is a legal question that precedes the technical one.
Historical data encodes historical bias
- Trained on who you hired beforeLearns who you hired before
- Protected fields removedProxies remain — test for them
- No outcome audit by groupYou cannot claim it is fair
- Evidence surfaced, human decidesDefensible
Historical data encodes historical bias
A model trained on who you hired before learns who you hired before, including the parts you would rather not repeat. Auditing outcomes across groups is not optional, and neither is testing for proxies once you remove the obvious fields.
Automate the logistics, not the judgement
- Scheduling, chasing, note structuringAutomate freely
- Drafting job descriptionsAutomate, with review
- Summarising against stated criteriaAutomate — surface evidence, no score
- Ranking or scoring candidatesCheck the law in every market first
- Rejecting without human reviewDon't
A defensible shape
The configuration we would build:
- Structured extraction from applications, with no scoring attached
- Explicit criteria defined by the hiring manager before applications open
- Evidence surfaced against each criterion, with the source quoted
- Human decides, every time, and the decision is recorded with reasons
- Regular outcome audits across demographic groups, reviewed by someone accountable