Skip to main content
skilder

governance

AI policy for employees: a usage policy template people follow

What an AI policy for employees should contain, a usage policy template you can adapt, and how to roll it out so employees use the approved tools instead of their own accounts.

Author: Nicolas Corod
  • #governance
  • #shadow-ai
  • #ai-governance
  • #roles
skilder mascot dressed as a lawyer, arms crossed and holding a gavel, beside the article title on a cream background with a light cyan disc

Most AI policy templates are correct. They list approved tools, ban confidential data in public chatbots and ask for a human check before anything leaves the building. What they rarely explain is why employees who have read the policy keep using their own accounts anyway.

The numbers suggest the gap is wide. In a Qualinsight survey of 1,030 professionals in French-speaking Switzerland, reported by PME in July 2026, 27% of people who use AI at work do so without approval from their company or its IT department, and 49% of respondents say their employer has set no clear rules on AI. This article covers what an AI policy for employees should contain, a usage policy template you can adapt, and how to roll it out so people follow it.

What is an AI policy for employees?

An AI policy for employees is a short internal document that states which AI tools staff may use, for which tasks, with which data, and who checks the result. It turns individual judgement calls into shared rules, so a sales manager and a payroll officer reach the same answer when they open an AI assistant.

It is not the same document as a governance framework, which sets how a company decides, approves and reviews AI use across the organisation; the guide to enterprise AI governance for leaders covers that layer. The employee policy is the part every person reads: one or two pages, written for the person at the keyboard, not for the board.

Different names cover the same document. An AI acceptable use policy, an AI usage policy and a company AI policy usually mean this one-page rulebook for staff.

Why do employees ignore most company AI policies?

Employees mostly route around AI rules because the approved option is slower or missing, not because they want to break them. A policy that bans without offering an alternative tends to move usage out of sight, where it becomes shadow AI: AI used for work through accounts the company does not provide or see.

The motive is usually efficiency. At a round table on AI in human resources at Salon RH 2026 in Lausanne, the panel described a presentation that took 30 minutes with Claude and three hours with the tool the company provided. Faced with that gap, a ban mostly teaches people to stop mentioning it.

The legal case for rules is still clear. David Raedler, lawyer and professor of employment law at the University of Lausanne, put it plainly in the same PME article: “The minimum is to set rules.” Under the General Data Protection Regulation (GDPR), the company remains responsible for the personal data its staff paste into a tool. In Switzerland, the Federal Act on Data Protection (FADP) allows fines of up to CHF 250,000 for certain intentional violations, imposed on the responsible individual rather than on the company.

So the question is not whether to have a policy. It is whether the policy makes the approved path easier than the hidden one. The shadow AI statistics for 2026 show how far the hidden path already reaches.

What should an AI acceptable use policy include?

An AI acceptable use policy needs seven sections: purpose and scope, approved tools, data rules, human review, prohibited uses, ownership and reporting, and review dates. Each one answers a question an employee will otherwise answer alone, usually in a hurry.

SectionThe question it answersCommon mistake
Purpose and scopeWho does this apply to, contractors included?Writing it for the IT team instead of every employee
Approved toolsWhich assistants may I use, with which account?Listing banned tools without naming an approved one
Data rulesWhat may I paste, and what never?”No confidential data” with no example of what that means
Human reviewWho checks AI output before it is used?Assuming the author checks, without saying when
Prohibited usesWhich decisions stay fully human?A long list nobody remembers
Ownership and reportingWho answers questions, and how do I report a mistake?No named owner, so every question goes to IT
ReviewWhen does this change?A policy written once and left for two years

Two sections decide whether the rest works. The data rules need three concrete tiers that an employee can apply in five seconds. The ownership section needs a named person, so the policy has someone to improve it.

AI usage policy template you can adapt

The template below is a starting point for a company of 50 to 1,000 people. Replace the bracketed parts, keep it to two pages, and remove any section that does not apply.

1. Purpose and scope. This policy sets how employees of [Company] use AI assistants for work. It applies to every employee, contractor and intern who uses an AI tool on [Company] tasks or data, on any device.

2. Approved tools. You may use [Claude / ChatGPT / Copilot, with the company account] for work. Do not use personal AI accounts for [Company] tasks. To request a new tool, write to [policy owner]; you will get an answer within [10 working days].

3. Data rules. Before you paste anything, check its tier:

  1. Public: published material, your own drafts without client or employee names. Allowed in approved tools.
  2. Internal: project documents, internal figures, client names. Allowed only in approved tools with the company account.
  3. Restricted: health data, salaries, evaluations, passwords, client data covered by a confidentiality agreement. Never in an AI tool unless [policy owner] has approved that specific use in writing.

4. Human review. You are responsible for anything you send, publish or decide with AI help. Check facts, figures and names before use. Say when a client deliverable was prepared with AI if the client contract requires it.

5. Prohibited uses. Do not use AI to make or recommend decisions on hiring, dismissal, pay or performance without a human decision on record. Do not use AI to create content that impersonates a real person.

6. Ownership and reporting. [Name, function] owns this policy. Report a mistake, such as restricted data pasted by accident, to [policy owner] within [24 hours]. Reporting early is expected and is not a fault in itself.

7. Existing use and review. If you already use AI tools that this policy does not list, declare them to [policy owner] by [date]. Use declared in this period carries no sanction and helps decide which tools to approve. This policy is reviewed every [six months] and after any incident.

How do you roll out a company AI policy that people follow?

Start from the use that already exists, write the rules with the people who use AI daily, and give them an approved path on day one. A policy announced from the top, with no alternative and no amnesty, mostly produces silence.

  1. Map current use without sanction. Ask each team which AI tools they use and for what. The Salon RH 2026 round table recommended an amnesty first, then a clear charter.
  2. Write the policy with the heaviest users. They know which rules are workable and which will be ignored by Thursday.
  3. Approve at least one tool before publishing. A policy that only forbids leaves people where they were.
  4. Name an owner per area, someone who answers questions and updates the rules as tools change.
  5. Explain the why in one paragraph. People follow a data rule more readily when they know which law or client contract it protects.
  6. Review after six months with the declared uses and reported incidents in hand.

How do you check that the policy holds in practice?

A written policy states intentions; checking it requires a record of what actually happened. Without that record, the six-month review relies on surveys and memory.

This is where tooling helps. Skilder works inside the assistants employees already use (Claude, ChatGPT or Copilot) and gives them business skills through roles configured from blueprints. A role bounds which skills and connected tools a team can reach, guardrails (in beta) check tool calls against rules the company writes, and skilder keeps an execution trace per person with an exportable audit log. The AI audit trail article shows what that record contains. To map your own policy onto your teams’ real use, book a free 30-minute exchange with an expert.

Takeaway: an AI policy works when the approved path is easier than the hidden one. Declare existing use without sanction, approve a tool before you forbid one, and name the person who keeps the rules current.

To turn this template into a policy your teams actually follow, book a free 30-minute exchange with an expert.

Frequently asked questions

What is an AI acceptable use policy?

An AI acceptable use policy is a short document that tells employees which AI tools they may use for work, which data they may put into them, who checks the output and which decisions stay human. It is usually one or two pages long and sits next to the IT, data protection and security policies. A usage policy only works if it names at least one approved tool, not just the banned ones.

Is it a legal requirement for an employer to have an AI policy?

In most European countries no law requires a document called an AI policy. The obligations come from existing rules: the General Data Protection Regulation in the European Union and its UK version, and the Federal Act on Data Protection in Switzerland, which allows fines of up to CHF 250,000 on the responsible individual for certain intentional violations. A written policy is the simplest way to show how staff are expected to meet them.

How do I control how my employees use ChatGPT?

Start by offering an approved account rather than a ban, then set clear data rules and a named owner for questions. Bans alone tend to push usage onto personal accounts. In a 2026 Qualinsight survey of professionals in French-speaking Switzerland, reported by PME, 27% of people using AI at work did so without approval from their company or its IT department. Visibility comes from an approved path plus a record of use.

How do I write an AI policy for employees?

Map the AI tools teams already use, without sanction, then write the rules with the heaviest users. Cover seven points: scope, approved tools, data tiers, human review, prohibited uses, a named owner and a review date. Keep it to two pages in plain language. Approve at least one tool before publishing, so the policy offers a way forward and not only a list of prohibitions.

What information should employees never put into an AI tool?

Restricted data should stay out of AI tools unless a named owner approves a specific use in writing: health information, salaries, performance evaluations, passwords and client data covered by a confidentiality agreement. Under the General Data Protection Regulation, the employer remains responsible for personal data its staff share with a tool. A three-tier rule (public, internal, restricted) is easier for employees to apply than a general ban on confidential data.

How often should a company AI policy be reviewed?

Review it at least every six months while AI tools change quickly, and after any incident or new tool approval. Each review should use real inputs: the tools employees declared, the questions the policy owner received and any mistakes reported. A policy that stays unchanged for two years usually describes tools nobody uses any more and misses the ones people rely on today.

Does Swiss data protection law apply when employees use ChatGPT or Claude?

Yes. When employees enter personal data into ChatGPT, Claude or Copilot for work, the company processes that data and the Federal Act on Data Protection applies, as it has since 1 September 2023. The company must know where the data goes and inform the people concerned. For certain intentional violations, the act allows fines of up to CHF 250,000 on the responsible individual rather than on the company.

Related articles