To create an AI policy for your company, start with the moment it is missing: Tuesday, 16:40, a project lead pastes a client list into a free chatbot to “tidy it up quickly”. Nobody told her not to. According to Bitkom, only 23 percent of German companies had rules for AI use in 20259. This guide shows how to write a short policy in six steps, with a skeleton you can copy, and how Superkind’s ready-made AI team turns the written rules into daily practice.
Key takeaways
Get a ready-made AI team - Superkind comes with an AI team for your policy (Policy Guide, Training Planner, Approval Guard, Tool Checker) that runs automatically. Start for free with unlimited users, or build your own.
Rules that are followed, not just filed - a line like “Quotes over EUR 10,000 need sign-off from management” becomes a rule Superkind applies in Outlook, Teams and Slack.
Keep control, keep data in Europe - Superkind is hosted in Europe, GDPR compliant, never trains on your data, and you decide which steps need approval.
Not mandatory, but practical - no law requires a written AI policy, but EU AI Act Art. 4, the GDPR and the works council make one the easiest way to stay on the safe side.
Two pages are enough - nine sections, from approved tools to the review date. Copy the skeleton below.
Your AI team for your AI policy
What is an AI policy?
An AI policy is a short internal rulebook that says which AI tools employees may use, which data must never go into them, which results need a person’s approval and who is responsible. It gives the team a clear yes instead of a vague maybe and documents what you do for AI literacy. Superkind is an example of an approved tool that also follows the policy: an AI team that does tasks in your programs, applies your approval rules and is hosted in Europe.
Is an AI policy mandatory?
- No general duty - no EU or German law requires a document called “AI policy”.
- EU AI Act Art. 4 - since 2 February 2025, companies that use AI must take measures for their staff’s AI literacy. Since the Digital Omnibus (in force 27 July 2026) the wording is “take measures to support”; the duty stays1, 2, 3.
- EU AI Act Art. 50 - since 2 August 2026, people must know when they interact with an AI system, and deepfakes or AI texts published to inform the public must be disclosed1, 3.
- GDPR - you must be able to show how personal data is processed (Art. 5 (2)), need a DPA with every processor (Art. 28) and sometimes a DPIA (Art. 35)5.
- Works council - where the company provides AI tools that can monitor performance, § 87 (1) no. 6 BetrVG usually gives it a say7, 8.
This guide is practical help, not legal advice. For your own case, ask your data protection officer or a lawyer.
How it works with Superkind: your AI team for your AI policy
You write to Superkind by email, in Teams, in Slack or in the web app, like a colleague, and Superkind does the task in your programs: via 1,700+ integrations, including Slack, Personio and SharePoint. It works with the tools your policy approves instead of replacing them, hosted in Europe and GDPR compliant.
With Superkind you don't start from zero: it comes with ready-made AI employees for your AI policy. They run automatically in the background, and you decide where they ask you first. Use them right away, adapt them or build your own.
Policy Guide
AI employee for questions about the AI policy
Nobody reads a twelve-page PDF before pasting a customer list into a chatbot. The Policy Guide answers questions in #ai-questions straight from your policy in SharePoint: which tool is approved, which data never goes in, who signs off. If the policy has no answer, it says so and tags the person responsible.
Training Planner
AI employee for AI literacy and onboarding
The EU AI Act expects you to support AI literacy, and nobody remembers who has done the training. When a new person is added in Personio, the Training Planner sends the policy and the training link, collects the confirmation and keeps the training list in Excel up to date. Every month you get the list of open items.
Approval Guard
AI employee for approval rules
Your policy says quotes over EUR 10,000 need sign-off from management. When a request comes in, the Approval Guard drafts the quote with your prices from HubSpot and stops right there if the total crosses the line. Smaller quotes go out on their own.
Tool Checker
AI employee for checking new AI tools
Someone found a new AI tool and wants to use it on Monday. The Tool Checker reads the vendor's privacy, DPA and pricing pages, notes where data is processed and whether inputs are used for training, and adds the result to your tool list. The decision stays with the person responsible for the policy.
When someone requests a new AI tool: check DPA, data location and training on inputs, add it to the tool list and ask me.
Reasoning ⌄Checked the transcription tool Mara requested. DPA available, processing location not stated on the vendor page, inputs used for training unless you opt out.
Every employee is ready to use right away. Or build your own on the platform.
Safe by default
An AI policy is only as safe as the tools it approves. With Superkind, the approved tool already runs in Europe, keeps your data out of model training and follows your approval rules.
- Hosted and processed in Europe
- GDPR compliant
- Aligned with the EU AI Act
- No training on your data
- DPA through the Trust Center
What you need
Before you start
- One person who owns the policy (owner, team lead, HR or the person responsible for data protection)
- A list of the AI tools your team already uses, including free and private accounts
- Your data protection officer, if you have one, and the works council, if there is one
- A place for the policy everyone can find: Notion, SharePoint or Google Drive
- A free Superkind account at app.superkind.ai, connected to your systems via 1,700+ integrations such as Google Drive, HubSpot and Personio (up to 150 hours of work free, then from EUR 50 per month for unlimited users, as of October 2026, see pricing)
Step by step with Superkind
Step 1: Find out which AI tools are in use
Ask the team openly and without blame. Superkind posts the question and collects the answers into one list.
Ask in #general which AI tools people use for work, for what and with which account. Collect the answers by Friday in a table.
You check the table for private accounts and tools that receive customer data. Those are the first rules you need.
Step 2: Decide approved tools and data classes
Sort every tool into approved, allowed with limits, or not allowed. Then define three data classes: public (allowed everywhere), internal (approved tools only) and personal or confidential (only tools with a DPA and processing in Europe)6.
Step 3: Write the policy from the skeleton
| Section | What goes in | Example |
|---|---|---|
| 1. Purpose and scope | Who it applies to, which tools | All staff, freelancers and interns |
| 2. Approved tools | List with allowed uses | Superkind for tasks with customer data |
| 3. Data that never goes in | Personal, health and secret data | No customer lists in unapproved tools |
| 4. Approval rules | Which results a person signs off | Quotes over EUR 10,000: management |
| 5. Responsibility | Owner, contact, escalation | Daniel owns it, questions in #ai-questions |
| 6. AI literacy (Art. 4) | Training per role, documented | Basics training in the first week |
| 7. New tools | DPA, data location, training check | Request first, use after approval |
| 8. Transparency (Art. 50) | Disclosure and labelling where it applies | Customer chat says it is AI |
| 9. Review cycle | Date and trigger for updates | Every six months, or on new tools |
Keep it to one or two pages. Bitkom’s guide on generative AI covers the legal background in depth10.
Write our AI policy from these nine sections and the tool list from Friday. Plain language, two pages max, save it in SharePoint under Team Handbook.
Superkind drafts the policy with your tools and examples. You read it, adjust and approve it.
Step 4: Turn the rules into rules Superkind follows
From now on: quotes over EUR 10,000 need sign-off from management. Never send personal data to tools outside our approved list.
Superkind confirms and applies the rules in every channel. Quotes above the limit wait for Approve or Reject; everything else runs on its own.
Step 5: Train the team and document it
Hold a short training per role: what the tools can do, where they make mistakes, which data stays out. The Bundesnetzagentur recommends documenting it4; the Training Planner keeps that list.
Step 6: Agree it, publish it, set a review date
Show the draft to your data protection officer and the works council, if you have one. Then pin it in Slack or Teams and set the next review date.
Set it up as a routine
On the first Monday of every quarter: list new AI tool requests, open trainings and questions the policy did not answer, and send me a summary.
- Every new hire - policy and training link go out from Personio as part of employee onboarding, confirmation lands in the training list.
- Every quarter - a short report: new tools, open trainings, questions without an answer.
- Every six months - you update the policy; Superkind keeps the new version in your Company Brain and answers from it.
As the rules grow, for example separate tool lists per department, you build your own app on the Superkind platform, on top of what Superkind already knows. It belongs to you.
Data protection and EU hosting
- Approval where you want it - you decide which steps wait for your OK before Superkind sends, books or changes something.
- Sovereign and independent - hosted and processed in Europe under European law, models from OpenAI, Anthropic, Mistral and Google processed in Europe, no lock-in to one model vendor.
- Computer use inside your setup - where a program has no integration, Superkind works in it via computer use, in the browser or desktop client, within your own workstation or RDS environment.
- DPA ready - sign the data processing agreement through the Trust Center.
Other ways
| Way | Effort | Error risk | Best for |
|---|---|---|---|
| Superkind | Low: write rules once, team follows them | Low: rules applied, approvals where needed | Teams that want rules enforced |
| Template from the web | Low to write, high to adapt | High: rules nobody checks | A first draft |
| Lawyer drafts it | Medium: fees and coordination | Low on paper, open in practice | Regulated industries, complex cases |
| Ban all AI | Low | High: use moves to private accounts | Rarely a lasting answer |
For the bigger governance picture, read our shadow AI governance playbook.
Common mistakes
- Banning without an alternative - a ban without an approved tool pushes use into private accounts.
- Twenty pages of legal text - nobody reads it. Two pages with examples get used.
- No owner - a policy without a named person goes stale within months.
- Training without proof - if you cannot show who was trained, Art. 4 is hard to demonstrate.
- Forgetting the works council - agreeing late costs more time than agreeing early.
6 tips for your AI policy
1. Start with a yes
Name the approved tool in the first paragraph. People follow rules when they know what they may use.
2. Use examples from your own team
“No client lists in free chatbots” is clearer than “no personal data in unapproved systems”.
3. Check the DPA before the first upload
For every tool with customer or employee data, check the DPA, where data is processed and whether inputs train the model.
4. Put approvals where people work
If management lives in Outlook, approvals go there. Superkind uses the same rules in Outlook, Teams and Slack.
5. Train by role
Sales needs different examples than HR; our AI literacy guide has the details.
6. Review after every new tool
Small updates beat a big rewrite once a year.
Conclusion
A good AI policy is short, names an approved tool and says who decides. Take the skeleton, write your two pages this week, then let Superkind apply the rules across Slack, Outlook and Personio via its 1,700+ integrations. Your first step: ask your team which AI tools they use today.
Frequently Asked Questions
There is no general legal duty to have a written AI policy. But the EU AI Act (Art. 4) expects every company that uses AI to take measures that support its staff's AI literacy, the GDPR requires you to prove how personal data is handled, and where a works council exists it often has a say. A short policy covers all three. This is not legal advice.
Nine sections are enough for most teams: purpose and scope, approved tools, data that never goes in, approval rules, who is responsible, AI literacy training, how new tools are checked (DPA, data location, training on inputs), transparency and labelling, and a review date.
If the company provides AI tools that can record how employees work, the works council usually has a right of co-determination under § 87 (1) no. 6 BetrVG. The Hamburg Labour Court ruled in January 2024 that allowing ChatGPT on private accounts did not trigger it, but that was a single case. Involve the works council early and agree the policy together.
The law does not prescribe a certificate or a fixed format. The Bundesnetzagentur recommends documenting your measures, for example who received which training and when. A list with name, date and topic is a good start.
Superkind turns the written rules into rules it follows: it answers policy questions in Slack or Teams, holds quotes over your limit until management approves, sends the policy and training to every new hire from Personio and checks new AI tools for you. It works in your programs through 1,700+ integrations, such as Outlook, Pipedrive and Zendesk, and natively via computer use where no integration exists.
Superkind is hosted and processed in Europe, GDPR compliant and aligned with the EU AI Act (Regulation (EU) 2024/1689). It does not train models on your data, and you sign the data processing agreement (DPA) through the Trust Center at trust.superkind.ai. AI models from OpenAI, Anthropic, Mistral and Google are processed in Europe.
Related Articles
- Shadow AI in the Mittelstand: The Governance Playbook
- AI Literacy for the Mittelstand: How to Implement Article 4 of the EU AI Act
- EU AI Act: The Omnibus Reprieve
- ChatGPT at Work: What’s Allowed, Forbidden, and Trained
- DPIA for AI Agents
- How to Use ChatGPT in Line with the GDPR
- How to Switch to European AI
- Sovereign AI Explained
Sources
- EUR-Lex - Regulation (EU) 2024/1689 (EU AI Act), Art. 4 and Art. 50
- EUR-Lex - Regulation (EU) 2026/1744 (Digital Omnibus on AI), published 24 July 2026
- Cuatrecasas - Digital Omnibus on AI has been published (Art. 4 and Art. 50 changes)
- Bundesnetzagentur - KI-Kompetenz (AI literacy under Art. 4)
- EUR-Lex - Regulation (EU) 2016/679 (GDPR), Art. 5, 28 and 35
- Datenschutzkonferenz (DSK) - Orientierungshilfe Künstliche Intelligenz und Datenschutz, 6 May 2024
- Gesetze im Internet - § 87 Betriebsverfassungsgesetz (BetrVG)
- Gleiss Lutz - Arbeitsgericht Hamburg zu ChatGPT (16 Jan 2024, 24 BVGa 1/24)
- Bitkom - Beschäftigte nutzen Schatten-KI (press release, 21 Oct 2025)
- Bitkom - Leitfaden Generative KI im Unternehmen
Send Superkind your first policy rule.
Start for free in Outlook, Teams or Slack. Or let us show you in 30 minutes how it runs in your systems.

