An AI acceptable use policy (AUP) is a short internal document on the use of generative AI at work. It sets which tools employees may use, which data they may enter, who reviews outputs and what happens when something breaks. The EU AI Act does not require one by that name, but your team already uses AI, so write the rules down.
What you’ll take from this
You will leave with a one-page policy you can adapt today and a plan to make your team follow it, supported if you like by the Certificate Program in Artificial Intelligence and Innovation from Founderz.
- Tell apart what the AI Act requires and what GDPR requires
- Decide which data may enter an AI tool
- Copy an AI acceptable use policy template and adapt it
- Roll it out in five steps without creating bureaucracy
What are AI acceptable use policies and why do AI policies matter?
An AI acceptable use policy is the internal rulebook for how employees work with tools such as ChatGPT or Microsoft 365 Copilot. It names approved tools, banned data, human review and owners. Without it, productivity gains can become a leak of personal data or an automated decision nobody can explain.
Use is already widespread. According to Eurostat (2025), 19.95% of EU enterprises used AI technologies in 2025, including 55.03% of large enterprises. The Work Trend Index from Microsoft and LinkedIn (2024) found that 78% of AI users at work bring their own tools, a habit security teams call shadow AI.
Are AI acceptable use policies legally required for the use of AI?
No regulation requires a document with that name, but two obligations make one necessary in practice. The EU AI Act asks organizations that use AI for human oversight and AI literacy measures, and GDPR requires you to protect personal data entering any tool.
Through Regulation (EU) 2026/1744, Article 4 of the AI Act now asks for measures that support AI literacy. Fines remain serious. In its original text, Article 99 of Regulation (EU) 2024/1689 sets fines for prohibited practices of up to 35 million euros or 7% of worldwide annual turnover, whichever is higher.
If an employee leaks personal data, Article 83 of the GDPR (Regulation (EU) 2016/679) sets two tiers: up to 10 million euros or 2% of worldwide annual turnover for lower-tier infringements, and up to 20 million euros or 4% for the most serious ones, whichever is higher.
Which laws govern generative AI at work in 2026?
Two EU laws apply at once, and one US framework helps. The AI Act regulates AI systems by risk, GDPR protects the personal data they process, and the NIST AI Risk Management Framework offers a voluntary way to manage these risks.
| Law | What it covers | What your policy should capture | Maximum fine |
|---|---|---|---|
| AI Act (Regulation 2024/1689) | AI systems by risk and the duties of those who use them | System inventory, human oversight, training and banned uses | 35 million euros or 7% (Article 99) |
| GDPR (Regulation 2016/679) | Processing of personal data | Banned data, vendor contracts and breach handling | 20 million euros or 4% (Article 83) |
Regulation (EU) 2026/1744 entered into force on 27 July 2026, so the 2 August 2026 date many guides cite no longer applies to high-risk systems. It moved those deadlines to 2 December 2027 and 2 August 2028, depending on system type. Confirm them on EUR-Lex before you put dates in your document.
If you use AI to screen candidates or evaluate performance, the AI Act treats it as high risk (Annex III), with obligations from 2 December 2027.
What should AI acceptable use policy templates include?
A clear policy has eight elements: scope, approved tools, allowed and banned data, human review, high-risk uses, transparency, incident handling and training. If one is missing, employees fill the gap themselves.
- Define scope, meaning who is covered and what counts as AI, including features built into existing software.
- Publish the approved tools list, with the account type allowed for each.
- Classify data and set which types never enter a tool.
- Require human review before any output is sent or published.
- Route high-risk uses through prior approval.
- State when to tell a client or colleague that content is AI-generated.
- Explain how to report an incident, and to whom.
- Tie the policy to training and a scheduled review.
What types of data can employees put into a generative AI tool?
It depends on the data’s classification and the vendor contract, not on what is convenient. Use this table as a starting point for your data handling rules and adjust it with your privacy or security lead.
| Data type | Example | Rule |
|---|---|---|
| Public | Press releases, website copy | Allowed in approved tools |
| Internal | Meeting notes, process drafts | Approved corporate accounts only |
| Confidential | Contracts, pricing, source code, trade secrets | Only if the vendor contract allows it |
| Personal and sensitive | Customer records, résumés, protected health information (PHI) under HIPAA | Prohibited without a prior data protection review |
What are acceptable uses of AI at work, and what should be banned?
Acceptable uses are low-risk tasks on approved tools with a human checking the result, such as summarizing internal notes, drafting public copy or brainstorming. Ban three things: personal or confidential data in free accounts, unreviewed output and decisions about people without approval.
Review AI-generated content for bias and accuracy before it reaches a client.
Artificial intelligence acceptable use policy template to copy (AI AUP)
This one-page template fits a small or midsize business. Copy it, then run it past Legal, IT and HR before you publish it.
AI acceptable use policy 1. Scope. Applies to all employees, interns and contractors, and to any tool with AI features. 2. Approved tools. Use only the tools on the official list kept by the AI owner. Personal accounts are not allowed. 3. Banned data. Never enter customer personal data, sensitive data, contracts or source code. 4. Human review. A person reviews and signs off every output before it is sent or published. 5. High-risk uses. Hiring, performance evaluation and automated decisions need prior approval. 6. Incidents. Report any misuse or leak to the AI owner the same day. 7. Review. The policy is reviewed every six months and after each serious incident.
How do you create and roll out an AI acceptable use policy?
Rollout takes five steps: inventory, draft, train, communicate and review. Each step has one owner and one output, so the policy reaches employees as a working routine instead of a PDF nobody opens, and it covers every approved AI tool and AI platform your teams use.
- Inventory tools and uses by asking each department, then rank them with our guide to identifying AI use cases in business.
- Name an AI owner who approves tools and high-risk uses, and draft the policy with Legal, IT and HR.
- Train the team with examples from their own work, because a rule nobody understands is not followed.
- Communicate the policy and open a channel to request exceptions.
- Revise the document based on the incidents and exceptions that arrive.
To structure that review as part of your AI governance framework, use the NIST AI Risk Management Framework (2023).
How do you control shadow AI and public AI tools at work?
Offer approved AI tools as an alternative to unapproved AI, with a corporate license and data terms reviewed by Legal, and explain why the free account is out. A ban without an alternative pushes use onto personal accounts, the habit the Microsoft study above describes, and unmanaged third-party accounts expose customer data, add a vulnerability and weaken data privacy.
How do you regulate AI agents in your AI governance policy?
Write one rule per AI agent stating what it can do without asking, what needs human approval and what is off limits, because an agent can act on systems rather than just draft text.
For automated decisions affecting people, require human review and the ability to explain the criteria. Article 22 of the GDPR also limits decisions based solely on automated processing that produce legal effects or similarly significantly affect the person.
Frequently asked questions
These are the questions that come up most when drafting acceptable use policies for AI.
How is an AI acceptable use policy different from an IT acceptable use policy?
An IT acceptable use policy protects systems, networks and access. An AI policy governs one specific technology: which tools, which data, which human review and which automated decisions. They complement each other, and the AI policy should point to the IT policy for passwords, access and breach handling.
Does the policy cover personal AI use on company devices?
Yes, it should. Say explicitly that personal AI accounts and apps are off limits for work data on any device, and cover AI features inside software employees already use.
How much AI use is acceptable?
None of the sources cited here sets a percentage cap on AI use. Set limits by data sensitivity, task risk and human review instead. A task is acceptable when an approved tool, permitted data and a named reviewer are all in place.
Further reading
These sources cover the legal text, the adoption figures and the risk framework cited above.
- Regulation (EU) 2024/1689, the AI Act, EUR-Lex
- Regulation (EU) 2016/679, GDPR, EUR-Lex
- Use of artificial intelligence in enterprises, Eurostat
- AI Risk Management Framework, NIST
To help your team follow the policy instead of only reading it, the Certificate Program in Artificial Intelligence and Innovation is an online program that lasts 3-4 months. It has 12 practical modules, covers cybersecurity challenges from the first module and teaches applying AI responsibly, with Microsoft 365 Copilot and Copilot Studio among the tools covered. You can request information or view the syllabus. All Founderz programs include the Founderz + Microsoft certification and official Microsoft badges.
This article is for informational and educational purposes only and is not legal advice or a substitute for professional judgment; always consult your attorney.
