Policy layout and proposed content. Company facts, operational commitments and legal applicability require approval before publication.
01. Purpose and scope
Use AI for defined support tasks such as organising evidence, preparing drafts and coordinating approved actions. Document which features are active, the relevant vendors and the data each task needs. This page describes design requirements, not a certification.
02. People remain accountable
Name the human owner for each workflow. Review information before candidate-impacting decisions. Provide the reviewer with source evidence and an escalation path. Do not equate a model output with an objective finding.
03. Fairness and assessment relevance
Use job-relevant criteria, monitor for harmful patterns and evaluate assessment quality. Avoid sensitive-trait inference and unsupported personality or emotion judgements. Plan accommodations and alternative formats where appropriate.
04. Transparency and correction
Inform candidates where AI is involved, explain the intended use and give them a way to raise inaccuracies or request human help. Keep meaningful explanations understandable to the people affected.
05. Data and model controls
Define permissions, retention, vendor processing terms and whether inputs may be used for model training. Test for unintended disclosure, prompt injection and inappropriate autonomous actions before enabling tools. Do not claim controls that have not been implemented and checked.
06. Monitoring and change
Assign owners for quality reviews, incidents and model or prompt changes. Retain proportionate records with a defined access and deletion policy. Review automation when roles, markets or data flows change.