Skip to content

Free templates

Two templates to start your AI governance

Editable Word documents you can adapt and use straight away. No account and no email address needed. Each one marks the details only you can fill in.

Word document · free

AI acceptable-use policy

A staff policy for choosing and using AI tools at work: approved tools, information that must not go in, uses that are not allowed, checking outputs, training and reporting problems.

What it covers

  • Approval of new AI tools and uses
  • Information that must not be entered
  • Prohibited uses, including Article 5 practices
  • Checking outputs and telling people (Article 50)
  • AI literacy training (Article 4)
  • Reporting problems, ownership and review
Download the template

Word document · free

AI supplier due-diligence questionnaire

12 questions to put to the supplier of an AI product or model, each with why it matters and the evidence to ask for, plus space to record answers, gaps and a decision.

What it covers

  • Accountability
  • System information
  • Data governance
  • Performance
  • Fundamental rights
  • Human oversight
  • Security
  • Monitoring and incidents
  • Change control
  • Transparency
  • Regulatory documentation
Download the template
Read the full text: AI acceptable-use policy

IMPORTANT REVIEW NOTICE

This is a free starting template from EU AI Fit. It is not legal advice, certification or proof that any obligation has been met. Replace every INFORMATION REQUIRED marker, remove what does not fit your organisation, and have the result reviewed by a competent adviser before you approve or rely on it.

1. Purpose and scope

This policy sets out how people working for [INFORMATION REQUIRED: organisation name] may choose and use AI tools and AI-enabled systems for work. It applies to employees, contractors and anyone else who uses AI on the organisation's behalf, whether the tool is built in-house, bought, or a general-purpose assistant used through a browser.

The EU AI Act asks providers and deployers of AI systems to take measures on the AI literacy of the people who operate or use those systems on their behalf (Article 4). This policy is one of those measures.

2. Principles

  • Use AI only for a purpose the organisation has approved and recorded.
  • A person stays accountable for every decision. Do not rely on an AI output in place of competent judgement where people's rights, safety, money or opportunities may be affected.
  • Check material outputs before using them, and say when content was produced with AI where people would expect to know.
  • Protect personal data, confidential information and intellectual property.
  • Raise concerns early. Reporting a problem in good faith is always the right thing to do.

3. Approved tools and how to request a new one

Approved AI tools: [INFORMATION REQUIRED: list the AI tools and AI-enabled systems approved for work, with the account type to use for each (for example a company account rather than a personal one).]

To use a new AI tool, or an approved tool for a new purpose, ask [INFORMATION REQUIRED: name the role or team that approves AI use] before you start. The request should say what the tool will be used for, who will use it, what information will go into it, who is affected by its outputs, and which supplier provides it.

New AI systems and material changes are recorded in the organisation's AI inventory before they are relied on.

4. Information you must not put into AI tools

Unless a tool has been specifically approved for it, do not enter:

  • personal data about customers, staff or other people;
  • special category data, such as health, ethnicity, religion or sexual orientation;
  • confidential business information, trade secrets, credentials or security details;
  • information covered by a contract, licence or legal privilege that restricts its use.

Organisation-specific restrictions: [INFORMATION REQUIRED: add any further restrictions, for example on client data, source code or regulated information.]

5. Uses that are not allowed

  • Any practice prohibited by Article 5 of the EU AI Act, such as social scoring, manipulative or deceptive techniques that cause significant harm, or inferring emotions in the workplace except for medical or safety reasons.
  • Making or materially shaping decisions about people, such as hiring, promotion, credit or access to services, without approval and a documented human review.
  • Presenting AI-generated content as the work of a named person, or publishing realistic synthetic images, audio or video of real people without approval and a clear disclosure.
  • Using AI to get round security controls, monitoring or this policy.

Further prohibited uses: [INFORMATION REQUIRED: add any uses your organisation does not allow.]

6. Checking outputs and telling people

  • Check facts, figures, citations and code produced with AI before you use them.
  • Where people interact with an AI system on the organisation's behalf, such as a chat assistant, make sure they are told they are dealing with AI unless it is obvious (Article 50).
  • Follow the approved instructions for each system, including any human oversight steps.

7. Training and competence

Everyone who uses AI for work completes the AI literacy training set for their role before using approved tools, and again when tools or rules change.

Training owner and refresh frequency: [INFORMATION REQUIRED: name who assigns and records AI literacy training, and how often it is refreshed.]

8. Reporting problems

Stop using a tool and report it straight away if you see harmful, discriminatory or misleading output, a possible data leak, a security issue, a loss of human control, or use outside the approved purpose.

Report to: [INFORMATION REQUIRED: name the contact or channel for AI incidents and how quickly they must be told.]

9. Ownership and review

Policy owner: [INFORMATION REQUIRED: name the policy owner.]

Approved by: [INFORMATION REQUIRED: name the approver and the approval date.]

Effective date: [INFORMATION REQUIRED: set the effective date.]

Review: at least once a year, and after any material change in the law, the organisation, its AI systems or its suppliers.

Read the full text: AI supplier due-diligence questionnaire

IMPORTANT REVIEW NOTICE

This is a free starting template from EU AI Fit. It is not legal advice, certification or proof that any obligation has been met. Replace every INFORMATION REQUIRED marker, remove what does not fit your organisation, and have the result reviewed by a competent adviser before you approve or rely on it.

How to use this questionnaire

Send it to the supplier of an AI product or model you buy or plan to buy, or work through it with them. It covers 12 areas of accountability, data, performance, oversight, security and regulatory information.

Answer No or Unknown where the supplier cannot show evidence yet. Each of those answers is a gap to follow up before you rely on the product, set a review date for, or accept as a documented risk. Use Not applicable only where the questionnaire allows it and your role and classification show the information is not needed.

Supplier: [INFORMATION REQUIRED: supplier legal name and contact]

Product or model, and version: [INFORMATION REQUIRED: the product, model and version covered]

How you use it: [INFORMATION REQUIRED: the purpose, users and people affected]

Reviewed by: [INFORMATION REQUIRED: reviewer name and date]

Next review date: [INFORMATION REQUIRED: when the answers will be checked again]

Questions

1. Accountability

Has the supplier identified the legal entity, responsible contact and its role in the AI value chain?

Why it matters: Confirm who provides the product or model, who supports it and which party carries each relevant responsibility.

Evidence to ask for: Contracting entity, governance contact and documented provider/deployer responsibility matrix.

Priority: High

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

2. System information

Has the supplier provided a current description of the product, model versions, intended uses and known limitations?

Why it matters: The information should be specific enough to compare the purchased service with its actual deployment.

Evidence to ask for: Product documentation, model card, release notes and limitations statement.

Priority: High

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

3. Data governance

Are data sources, permitted data uses, retention, locations and subprocessors sufficiently documented?

Why it matters: Include customer inputs, service improvement or training use, deletion and international transfers.

Evidence to ask for: Data processing agreement, retention schedule, subprocessor list and data-flow description.

Priority: High

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

4. Data governance

Where relevant, has the supplier documented training-data provenance, quality and rights controls?

Why it matters: Use not applicable only where the purchased service does not require this information for the organisation's role and use.

Evidence to ask for: Training-data summary, provenance statement, copyright policy and quality controls.

Priority: Medium

Answer: Yes / No / Unknown / Not applicable

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

5. Performance

Are performance claims, evaluation methods, failure modes and appropriate-use boundaries documented?

Why it matters: Look for evidence that reflects the intended context, not only a general marketing benchmark.

Evidence to ask for: Evaluation report, test dataset description, error analysis and acceptable-use guidance.

Priority: High

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

6. Fundamental rights

Has the supplier addressed foreseeable bias, accessibility and impacts on affected people?

Why it matters: The depth should be proportionate to the decisions and people affected by the deployment.

Evidence to ask for: Bias or fairness evaluation, affected-group analysis, accessibility evidence and mitigation plan.

Priority: Medium

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

7. Human oversight

Does the supplier explain the human oversight, override, escalation and safe-stop capabilities available to the customer?

Why it matters: Confirm the controls exist in the supplied version and can be used by the client's operators.

Evidence to ask for: Instructions for use, role permissions, override test and escalation procedure.

Priority: High

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

8. Security

Are security, resilience, vulnerability handling and access controls supported by current evidence?

Why it matters: Reuse established security assurance where it covers the actual AI service and hosting arrangement.

Evidence to ask for: Security report or certification, penetration-test summary, vulnerability policy and resilience evidence.

Priority: High

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

9. Monitoring and incidents

Are monitoring responsibilities, incident notification and investigation cooperation contractually clear?

Why it matters: Confirm notification timing, information supplied and regulator/customer cooperation expectations.

Evidence to ask for: Incident clause, monitoring specification, service levels and tested escalation contacts.

Priority: High

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

10. Change control

Must the supplier notify the customer of material model, data, capability, terms or subprocessor changes?

Why it matters: Notice should arrive early enough for the client to reassess exposure, risks and controls.

Evidence to ask for: Change-notification clause, release process, version history and deprecation policy.

Priority: High

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

11. Transparency

Does the supplier provide the information needed for user notices, instructions and downstream transparency duties?

Why it matters: Check the information supports the client's real interface, outputs and affected users.

Evidence to ask for: Transparency information, user-notice guidance, synthetic-content marking specification and instructions.

Priority: Medium

Answer: Yes / No / Unknown

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

12. Regulatory documentation

Where applicable, are conformity, registration and technical documents available and tied to the supplied version?

Why it matters: Use not applicable only after the role and classification decision shows these documents are not required for this relationship.

Evidence to ask for: Declaration of conformity, CE information, registration evidence and version-specific technical documentation.

Priority: High

Answer: Yes / No / Unknown / Not applicable

Notes and evidence received: [INFORMATION REQUIRED: record the supplier's answer, the documents received and any follow-up.]

Outcome

Overall decision: [INFORMATION REQUIRED: approve, approve with conditions, or do not use, with the reason.]

Conditions and actions: [INFORMATION REQUIRED: list each follow-up action, its owner and its due date.]

Next step

Put the templates to work

Find out what may apply to you

The free exposure check screens one AI use case in a few questions and shows which duties may apply and what to do next.

Run the free check

Run supplier reviews in one place

In the workspace, the same supplier questions go out by secure single-use link, and every answer, gap and action stays on the supplier’s record.

See supplier assurance

Last reviewed 16 November 2026. These templates are a starting point, not legal advice or certification. Have the finished documents reviewed by a competent adviser before you rely on them.