An AI acceptable use policy is a company's rulebook for everyday AI use: which tools staff may use, what data may go into them, when a person must check the output, how AI use is disclosed to customers, what training staff receive, and how problems are reported. The template below gives model wording for each clause, ready to copy and adapt.
What a company AI policy needs to cover
Most organisations already use AI daily, approved or not: chat assistants, meeting summaries, coding tools and AI features switched on inside existing software. An acceptable use policy turns that into a bounded practice, and stays short so that staff remember it.
| Clause | Why it is there | Rule it helps you meet |
|---|---|---|
| Purpose, scope and roles | Says who is bound and who decides | Internal accountability |
| Approved and prohibited tools | Keeps company data out of unvetted services | GDPR Art. 28 processor terms |
| Data classification | Maps each class of information to the tools allowed to receive it | GDPR Art. 5(1)(c) and Art. 9 |
| Human review | Keeps a person responsible for decisions and published work | GDPR Art. 22 |
| Disclosure | Tells people when they deal with AI | EU AI Act Art. 50, applicable since 2 August 2026 |
| AI literacy | Gives staff the knowledge to use AI safely | EU AI Act Art. 4, applicable since 2 February 2025 |
| IP and confidentiality | Protects trade secrets, client material and third-party rights | Contracts and confidentiality duties |
| Incident reporting | Gets leaks and errors to the right team fast | GDPR Art. 33 breach notification |
| Exceptions, enforcement, review | Keeps the policy credible and current | Internal governance |
The AI acceptable use policy template
Copy each block into your own policy format. Replace the bracketed placeholders, such as [Company] and [AI governance lead], and delete clauses that do not apply. This is a starting draft for your legal, privacy, security and HR teams to review, not legal advice.
1. Purpose, scope and roles
1. PURPOSE, SCOPE AND ROLES
1.1 Purpose. This policy sets the rules for using artificial
intelligence (AI) tools at [Company], so that staff can use AI
productively while protecting customers, colleagues, confidential
information and personal data.
1.2 Scope. It applies to all employees, contractors, temporary staff
and anyone else who uses AI tools for [Company] work, on any
device, including AI features built into software we already
license.
1.3 Definition. An "AI tool" is any system that generates text, code,
images, audio, video, summaries or recommendations from a prompt
or other input, including chat and coding assistants, meeting
summary features and AI agents that act in other systems.
1.4 Roles.
- Policy owner: [AI governance lead / CIO] maintains this policy
and the Approved AI Tools Register.
- AI review group: [members from IT security, privacy (the data
protection officer), legal and the business] approves new
tools, new uses and exceptions.
- Managers make sure their teams know and follow this policy.
- Users follow this policy, complete the required training and
report concerns.2. Approved and prohibited AI tools
2. APPROVED AND PROHIBITED AI TOOLS
2.1 Approved tools only. Use only the AI tools listed in the Approved
AI Tools Register, with the accounts and settings stated there.
Do not use personal AI accounts for [Company] work.
2.2 Approval. A new AI tool, or a new use of an approved tool, needs
approval from the AI review group before use. The review covers
security; data protection, including a data processing agreement
and, where needed, a data protection impact assessment; where
data is processed and stored; whether the vendor trains models on
our data; retention; and contract terms.
2.3 Prohibited uses. Unless the register says otherwise, you must not:
- use free or consumer versions of AI tools for [Company] work;
- install AI browser extensions, plug-ins or agents that can read
company data;
- connect an AI tool to company email, files, code repositories
or business systems without approval;
- use AI to make decisions about a person's employment, credit,
insurance, access to services or legal rights without the
review in section 4;
- use AI to impersonate a real person or to create misleading
content about real people or organisations;
- try to bypass the safety controls of an AI tool or our own
security controls.
2.4 Embedded AI features. AI features that vendors add to software we
already use count as new tools. They stay switched off until the
AI review group has reviewed them.For the review in clause 2.2, our 13-point GDPR check for AI tools lists the questions to send each vendor.
3. Data classification: what may go into which tool
| Data class | Examples | Approved external AI tools | Internal AI on [Company] infrastructure |
|---|---|---|---|
| Public | Published marketing copy, public web content | Allowed | Allowed |
| Internal | Internal procedures, non-sensitive project notes | Only tools approved for Internal data | Allowed |
| Confidential | Customer records, contracts, financials, unreleased plans, source code | Only tools approved for Confidential data, with training off and the agreed processing region | Allowed |
| Restricted | Special-category personal data, privileged material, credentials, security details | Never | Only systems approved for Restricted data |
3. DATA RULES: WHAT MAY GO INTO WHICH TOOL
3.1 Classify before you paste. Before entering information into an
AI tool, check its class under the [Information Classification
Policy] and the table above. If unsure, treat it as Restricted.
3.2 Never into an external AI tool. The following must never be
entered into an AI tool hosted outside [Company]'s infrastructure:
- passwords, API keys, tokens or other credentials;
- special categories of personal data (health, genetic or
biometric data, racial or ethnic origin, political opinions,
religious or philosophical beliefs, trade union membership,
sex life or sexual orientation) and data about criminal
convictions or offences;
- legally privileged material, and client material that the
client's contract does not allow us to process this way;
- details of unpatched vulnerabilities or open security
incidents;
- data that a contract, licence or law requires us to keep in a
specific location or system.
3.3 Minimise. Share only what the task needs. Remove names and other
identifiers whenever the task can be done without them.
3.4 Personal data. Use personal data in AI tools only for purposes
already covered by our records of processing and privacy
notices, and only in tools covered by a data processing
agreement. A new purpose needs the privacy team's approval first.4. Human review of AI output
4. HUMAN REVIEW OF AI OUTPUT
4.1 You own the result. AI output is a draft. The person who uses it
is responsible for checking facts, figures, sources, code and
tone before relying on it or sharing it.
4.2 Mandatory review. A qualified person must review and approve AI
output before it is:
- sent to a customer, regulator, court or the public;
- used in a contract, financial report or legal advice, or in a
medical or safety-relevant context;
- merged into production code or used to change production
systems.
4.3 Decisions about people. AI must not be the sole basis for a
decision with legal or similarly significant effects on a
person, such as hiring, dismissal, credit or access to a service.
A person with authority to change the outcome makes the decision
and records the reasons.
4.4 AI agents. Agents may act (send messages, update records, start
payments, change systems) only within approved permissions.
Actions with financial, legal or customer impact need human
approval before they run.5. Disclosure to customers and the public
5. DISCLOSURE TO CUSTOMERS AND THE PUBLIC
5.1 AI interactions. Any AI system that talks or writes to customers
or the public on [Company]'s behalf must tell people they are
dealing with AI, clearly and at the latest at the first
interaction, unless that is obvious from the context.
5.2 Synthetic media. AI-generated or AI-altered images, audio or
video that show real people, places or events in a way that could
pass as authentic must be labelled as AI-generated.
5.3 Published text. Text generated by AI and published to inform the
public on matters of public interest must be labelled as such,
unless a person has reviewed it and [Company] takes editorial
responsibility for it.
5.4 Contracts first. Where a contract or client instruction requires
us to disclose or limit AI use, that requirement applies in
addition to this policy.These clauses follow Article 50 of the EU AI Act, which has applied since 2 August 2026. The Commission published guidelines on it on 20 July 2026, and our Article 50 explainer covers who counts as the provider of an in-house assistant.
6. AI literacy and training
6. AI LITERACY AND TRAINING
6.1 Required training. Everyone in scope completes [Company]'s AI
training before using approved AI tools for work, with a
refresher every [12] months. Content differs by role: general
users, people who build or configure AI tools, and people who
review AI-assisted decisions.
6.2 Content. How the approved tools work and fail, including
inaccurate or invented output; the data rules in section 3; the
review rules in section 4; disclosure duties; and how to report
incidents.
6.3 Records. The [learning team] keeps an internal record of who
completed which training and when.Article 4 of the EU AI Act has applied since 2 February 2025. As amended by the Digital Omnibus on AI, Regulation (EU) 2026/1744, it asks providers and deployers to take measures that support the AI literacy of their staff, without guaranteeing a set level. The Commission’s Q&A says no certificate is needed and that an internal record of training can be kept.
7. Intellectual property and confidentiality
7. INTELLECTUAL PROPERTY AND CONFIDENTIALITY
7.1 Confidentiality applies. Duties of confidentiality in employment
contracts, client contracts and non-disclosure agreements apply
to AI tools exactly as they apply to any other third party.
7.2 Trade secrets. Source code, algorithms, formulas, pricing and
strategy documents go only into tools approved for Confidential
data.
7.3 Third-party rights. Do not use AI to reproduce copyrighted
material, trademarks or a person's likeness without permission,
and check generated code for licence obligations before use.
7.4 Ownership. Work created with AI tools for [Company] belongs to
[Company] to the extent the law and the tool's terms allow.
Record significant AI contributions to deliverables where a
client or contract requires it.8. Incident reporting
8. INCIDENT REPORTING
8.1 What to report. Report straight away if you:
- entered data into an AI tool that this policy does not allow;
- receive AI output that is harmful, discriminatory, defamatory
or reveals someone else's data;
- notice an AI agent acting outside its permissions;
- find an unapproved AI tool in use with company data.
8.2 How. Report to [security contact or ticket queue] within
[24 hours]. Do not delete evidence while trying to fix the
problem.
8.3 Personal data breaches. The privacy team decides whether an
incident is a personal data breach. Under Article 33 of the
GDPR, a breach must be notified to the supervisory authority
without undue delay and, where feasible, within 72 hours, unless
it is unlikely to result in a risk to people's rights and
freedoms.9. Exceptions, enforcement and review
9. EXCEPTIONS, ENFORCEMENT AND REVIEW
9.1 Exceptions. Send exception requests to the AI review group in
writing, with the business reason, the data involved and the
safeguards proposed. Approved exceptions are time-limited,
recorded in the register and reviewed when they expire.
9.2 Enforcement. Breaches of this policy are handled under
[Company]'s disciplinary procedure and can lead to loss of access
to AI tools. For contractors, contractual remedies apply.
9.3 Review. The policy owner reviews this policy at least every [12]
months, and sooner when laws, regulator guidance, approved tools
or our risk profile change.
9.4 Version control. Version [x.y], approved by [name, role] on
[date]. Next review due: [date].How to roll out the policy
- Find what is already in use. Survey teams and check network logs and expense claims; the result is your first register. An AI inventory and shadow AI discovery exercise gives this step structure.
- Approve a usable default first. Configure at least one tool for Internal and Confidential data (training off, retention set, region chosen, access by role) before publishing, so the policy offers a working route alongside its bans.
- Agree the data classes with privacy and security, and fill in the section 3 table with real tool names.
- Publish with a one-page summary, because most staff will read only that.
- Train by role, starting with heavy users and reviewers of AI-assisted decisions, and keep completion records.
- Make reporting easy by putting the section 8 channel where people already ask for IT help.
- Review after 90 days, using exception requests, incidents and tool requests to adjust the register and the policy.
One-page AI policy checklist
- A named policy owner and an AI review group with security, privacy and legal members
- An Approved AI Tools Register listing accounts, settings and permitted data classes
- Personal and consumer AI accounts banned for company work
- Four data classes mapped to tools, plus a never-share list for external AI
- Data processing agreements, and DPIAs where needed, for tools that receive personal data
- Human review before external, legal, financial, safety-relevant or production use
- No solely automated decisions with legal or similarly significant effects
- AI disclosure at the first interaction for customer-facing chatbots and agents
- Labels for AI-generated images, audio and video of real people or events
- Role-based AI training with completion records
- Rules on confidentiality, client contracts, copyright and code licences
- An incident route with a reporting deadline and a link to breach handling
- A written exceptions process with expiry dates
- An annual review date and triggers for an earlier review
Adapting the template for law firms, healthcare and small teams
- Law firms and professional services. Engagement terms often decide which tools may see client material, so let client data go only into tools the engagement allows, and keep privileged material Restricted.
- Healthcare and life sciences. Patient data is special-category data under Article 9 of the GDPR, so it stays Restricted.
- Small businesses. A two-person review group and a shared-document register are enough, but keep the data table and the never-share list intact, since they carry most of the protection.
Draft it with your own details
The AI governance policy generator drafts AI use policies, RACI matrices, approval lifecycles and risk-appetite statements grounded in EU AI Act Article 26 and ISO/IEC 42001, on infrastructure you control. Once the policy sits beside your other policies, the policy review agent reads the set together and reports contradictions, gaps and overdue reviews.
For section 6, the AI compliance training agent delivers role-adaptive Article 4 training with scenario questions, and for section 5 the transparency notice generator drafts plain-language disclosure wording. The Academy lesson on building a governance framework and RACI covers the review group and approval flow behind the policy.
How VDF AI fits
A policy holds when the approved route is easier than the workaround. VDF AI gives staff assistants and agents that run on your own infrastructure, on-premises, in a private cloud or air-gapped, so Confidential and Restricted data can be used with AI without leaving it.
Role-based access control, included on every plan, decides which roles can use which agents, tools and workspaces. Consequential agent steps can wait at a human approval gate, and every prompt, tool call, retrieval and response is recorded with the user and time, which turns sections 2 to 4 of the policy into enforced settings rather than reminders.