Responsible AI hiring policy your team can actually use.

1. Purpose

[Company] uses AI and automation in hiring only to reduce menial work, surface decision-relevant evidence, and improve consistency — never to replace human judgment for selection, rejection, or other irreversible candidate decisions.

2. Scope

This policy applies to recruiters, hiring managers, interviewers, people operations, vendors, and any system that processes candidate data for [Company] roles. It covers generative AI, ranking suggestions, summarization, scheduling automation, chatbots, and third-party AI tools used in hiring.

3. Guiding principles

• Human dignity: candidates are whole people, not scores, keyword matches, or prestige bundles.
• Human-in-command: a named human remains accountable for consequential decisions.
• Fair process: candidates for the same role receive the same ordered process unless an audited exception is documented.
• Truth & evidence: AI outputs must be inspectable against candidate-provided or process-generated evidence.
• Minimal data: collect and process only what is needed for a fair hiring decision.
• Transparency: candidates and internal stakeholders can understand when AI assisted a workflow at a high level appropriate to the context.

4. Allowed uses

• Drafting job descriptions and internal process templates for human edit and approval.
• Summarizing candidate-provided materials with citations or excerpts for human review.
• Suggesting rubric scores or interview notes drafts that a human must accept, edit, or reject.
• Scheduling logistics, reminders, and status notifications.
• Email sequence drafts and personalization tokens under human configuration and review rules.
• Search and retrieval of evidence (for example ownership or values examples) without final ranking authority.

5. Prohibited uses

• Autonomous rejection, selection, or final ranking of candidates by AI.
• Opaque scoring that cannot be explained with decision-relevant evidence.
• Using protected-class proxies, demographic inference, or prestige theater as fitness signals.
• Silent overrides of fair stage order without an accountable human and documented reason.
• Feeding sensitive candidate data into unapproved consumer AI tools outside company policy.
• Making irreversible hiring or adverse decisions without a human decision record.

6. Roles and accountability

• Policy owner: [Name / role] maintains this document and annual review.
• System owners: [ATS admin / People Ops] control which AI features are enabled per tool and tier.
• Hiring managers & recruiters: remain accountable for stage advancement, rejection explanations, and final recommendations.
• Interviewers: score from evidence; treat AI drafts as optional assistance, never as the decision.
• Legal / compliance: reviews material changes, vendor AI, and jurisdiction requirements.

7. Human review requirements

• Any AI suggestion that could affect advancement or rejection requires human review before action.
• Adverse decisions require a human-authored candidate-facing explanation and internal decision category.
• AI-assisted summaries must be checkable against source evidence when used in a hiring decision meeting.
• Feature toggles default to the least automated setting that still supports the team until policy and training are complete.

8. Data protection

• Candidate data is used only for legitimate hiring purposes and retained per [retention schedule].
• Access is limited to people with a need to know for the role.
• Candidates may submit access, correction, deletion, or export requests through [privacy process / link].
• Vendor AI tools must meet [Company] security and data-processing standards before enablement.

9. Fairness and monitoring

• Review funnel drop-off, stage time, override rates, and adverse-decision categories on a regular cadence.
• Investigate unexplained exclusion patterns as process defects, not as inevitable outcomes.
• Train hiring teams on structured interviews, human-signal rubrics, and AI limitations.
• Document material model/prompt/vendor changes that affect hiring assistance.

10. Exceptions and escalations

Exceptions to this policy require written approval from [policy owner] and, where relevant, legal. Escalations about candidate harm, bias concerns, or AI misuse go to [escalation path] within [SLA].

11. Review cadence

This policy is reviewed at least annually, and whenever [Company] enables a new AI hiring vendor, materially expands automation, or receives a significant fairness or privacy incident. Version: [0.1] · Effective: [date] · Last reviewed: [date].