Skip to content
Free Employee Handbook Template

Guide, 10 Nov 2026, read 10 Nov 2026

An AI use policy for an HR team lists the tools, the data, the reviewer and the audit date

See the rankingAll guides

Start with a register, not a rule

Before writing rules, list every AI tool the People team uses or plans to use, including features inside existing systems: résumé screening, interview scheduling, drafting job ads, policy chat tools, survey analysis and performance summaries. For each, record the purpose, the owner, the data it receives and where that data goes.

This register is the backbone of the policy. A rule that applies to a tool nobody has listed will not be followed, and a register is what an auditor, a candidate or a regulator asks to see first.

Know which uses are regulated

Hiring and worker management tools carry legal duties. The EU AI Act lists AI used to recruit or select people, filter applications, evaluate candidates or make decisions on work terms as high-risk, and the European Commission page gives 2 December 2027 as the date the rules for those systems apply. The ban on prohibited practices and the AI literacy duty have applied since 2 February 2025. Providers of high-risk systems must also have a quality management system under Article 17.

In New York City, Local Law 144 bars employers from using an automated employment decision tool unless it has been subject to a bias audit within one year, a summary is public and notices have been given. Other states and cities are adding rules, so record the location of each candidate pool.

Set the rules that matter

  • Approved tools and purposes: name them, and say that anything not listed needs approval from the policy owner.
  • Data: no special category data, health data or candidate files in tools that have not been approved for it; follow the UK GDPR and the privacy notice for staff.
  • Human review: a named person reviews every output that affects a hiring, pay, promotion or dismissal decision, and records the reason for the final choice.
  • Notice: candidates and staff are told when an automated tool is used and how to ask for another route.
  • Testing: each tool is tested for adverse impact before use and at least once a year, with the result kept.
  • Incidents: who is told, and how quickly, when a tool produces a wrong or biased result.

Decide how decisions are explained

A rule that says a human decides is only true if the human can explain the decision. Ask reviewers to write the reason in their own words. If a candidate or an employee asks why, the answer should not be 'the tool said so'. Under UK GDPR, people have rights to see data held about them, and a request for access can reach notes and scores.

For tools that draft text, such as job ads or policy summaries, require a read-through by the owner before publication and a check for claims the company cannot support.

Give the policy an owner and a review rhythm

Name the owner, a deputy and a review cycle. Review the register every quarter in the first year, since tools change quickly, and once a year after that. A voluntary framework such as ISO/IEC 42001 sets out how an organization governs its AI with an owner, risk review and internal audit; a small team can borrow the shape without certifying.

Train everyone who uses a tool in the rules and keep a record. The AI literacy duty in the EU is a reason to do so, and it is a good idea everywhere.

What to do in the first month

  • Week one: build the register of tools and owners.
  • Week two: classify each use as hiring, people decisions or general drafting, and flag the regulated ones.
  • Week three: write the policy on one page and agree it with legal and the head of People.
  • Week four: train the team, publish the register and set the first review date.

Sources

  1. AI Act, European Commission, read 10 Nov 2026
  2. Regulation (EU) 2024/1689, EUR-Lex, read 10 Nov 2026
  3. Automated Employment Decision Tools, NYC DCWP, read 10 Nov 2026
  4. UK GDPR Article 15, legislation.gov.uk, read 10 Nov 2026

Last reviewed 10 Nov 2026.