How to Write a Generative AI Policy for Your Business

How to write a workplace generative AI policy that people will follow: approved tools, rules for data, human review, intellectual property, disclosure and alignment with GDPR and the EU AI Act.

Abstract illustration of connected nodes representing artificial intelligence systems

Key points

  • Staff are already using generative AI; a policy channels that use instead of pretending to prevent it.
  • The core rules concern what data may be entered, which tools are approved and who checks the output.
  • Legal anchors include data protection law, confidentiality, copyright and, in the EU, the AI Act’s literacy and transparency duties.
  • A short policy with examples, backed by training, works better than a long one nobody reads.

Generative AI tools draft emails, summarise documents, write code and analyse spreadsheets. Their adoption has been led by employees, often using personal accounts, ahead of any corporate decision. That creates real exposure: confidential information pasted into a public service, invented facts sent to a client, code of uncertain origin added to a product. A clear policy lets the organisation capture the productivity gain while controlling those risks.

What the policy needs to cover

1. Scope and approved tools

State which tools are approved for work and under which licence. Enterprise versions typically commit by contract not to use customer inputs to train models, offer administrative controls and provide a data processing agreement. Consumer versions often do not. Name the approved tools, explain how to request a new one, and prohibit work use of personal accounts.

2. Data classification rules

Type of informationPublic or consumer AI toolApproved enterprise tool
Public informationPermittedPermitted
Internal, non-sensitiveNot permittedPermitted
Personal dataNot permittedOnly for approved use cases after a data protection review
Client confidential or contractually restrictedNot permittedOnly where the client contract allows it
Special category data, credentials, trade secrets, unreleased financial resultsNot permittedNot permitted without specific sign-off
An illustrative matrix. Adapt the categories to your own classification scheme.

3. Human review and accountability

Generative models can produce fluent text that is wrong, including invented citations, figures and case law. The policy should state that the employee who uses the output is responsible for it, that facts, calculations and references must be verified against sources, and that AI output is never the sole basis for decisions about individuals, such as hiring, discipline, credit or pricing.

4. Intellectual property

  • Output may resemble existing copyrighted material. Check before publishing or shipping.
  • Purely machine-generated content may not attract copyright protection in some jurisdictions, which matters for work delivered to clients as original.
  • For software, apply the same licence scanning and code review to AI-generated code as to any third-party code.
  • Do not upload third-party material you have no right to share.

5. Disclosure

Decide when AI use must be disclosed: to clients where a contract or professional rule requires it, to the public for chatbots and synthetic media, and internally where material work product was machine-generated. From August 2026 the EU AI Act requires that people are informed when they interact with an AI system and that deepfakes are labelled.

6. Prohibited uses

List them plainly: generating discriminatory or harassing content, impersonating real people, circumventing security controls, making automated decisions about individuals without approval, and any practice banned by the AI Act.

7. Security

Cover browser extensions and plug-ins that read page content, AI meeting assistants that record calls, and agents with access to email or files. Each needs approval, since they can exfiltrate data beyond what the user types.

The legal anchors

  • Data protection. Entering personal data into an AI tool is processing under the GDPR. You need a lawful basis, a processor agreement and, for higher-risk uses, a data protection impact assessment. The UK Information Commissioner’s Office has published guidance on AI and data protection.
  • AI literacy. Article 4 of the AI Act, applicable since February 2025, requires providers and deployers to ensure a sufficient level of AI literacy among staff using AI systems.
  • Confidentiality. Contracts and professional duties may forbid sharing client information with third-party services.
  • Employment law. In several EU countries, works councils have information or co-determination rights over the introduction of tools capable of monitoring employees.
  • Sector rules. Financial services, healthcare and legal regulators have issued their own expectations.

Making it work

  1. Involve users. Draft with people from the teams that use the tools most.
  2. Give approved alternatives. A ban without a sanctioned tool drives use underground.
  3. Train with examples drawn from real tasks, including prompts that would breach the policy.
  4. Keep a register of approved tools and use cases; it doubles as the inventory the AI Act expects.
  5. Review every six months. Tools and law are changing quickly.
  6. Assign an owner, with a simple route for questions and incident reports.

Frequently asked questions

Should we ban generative AI at work?

Outright bans are hard to enforce and forgo real benefits. Most organisations do better by approving specific tools with clear data rules.

Do vendors train their models on our data?

It depends on the product and the contract. Check the terms of the specific plan; enterprise agreements commonly exclude training on customer data.

Who is liable for errors in AI-generated work?

The organisation and the professional who delivered it. Using an AI tool does not transfer responsibility.

The bottom line

A generative AI policy should fit on two or three pages and answer three questions for any employee: which tools may I use, what may I put into them, and what must I check before I rely on what comes out. For practical uses in a finance function, see AI in finance teams.

Important: This article is general information, not legal advice. Adapt any policy to your contracts, sector rules and the law of the countries where you operate.

Sources and further reading

Scroll to Top