Which rules apply to AI in business?
Two EU regulations are decisive for most companies:
- The General Data Protection Regulation (GDPR) applies whenever personal data is processed: names, email addresses, customer data or information about employees. It doesn't matter whether a person or an AI processes it.
- The EU AI Act (Regulation (EU) 2024/1689) governs AI systems themselves. It has been in force since 1 August 2024 and applies in stages.
Depending on the application, further rules come into play, for example on copyright, on employment law for AI in HR, or industry-specific requirements. This article provides an overview and is not legal advice.
What does the GDPR require when using AI?

- Legal basis: Any processing of personal data needs a basis under Art. 6 GDPR, such as a contract, a legitimate interest or consent.
- Data minimization: The AI only receives the data it needs for the task. Names and identifiers can often be removed or replaced beforehand.
- Data processing agreement: If you use an external AI service, it processes data on your behalf. This requires a contract under Art. 28 GDPR.
- Data transfers: If the provider is based outside the EU, for example in the US, the transfer must be safeguarded. Many providers now offer processing within the EU.
- Transparency: Data subjects must learn that and why their data is processed. This belongs in your privacy policy.
- No automated decisions with significant effects: Decisions that significantly affect people, such as on a job application, generally must not be made by a machine alone (Art. 22 GDPR).
- Impact assessment for high risk: If the AI processes a lot of sensitive data or evaluates people, a data protection impact assessment under Art. 35 GDPR is often required.
Important in practice: check whether the provider uses your inputs to train its models. For company data, this should be excluded by contract.
How does the EU AI Act classify AI systems?
| Risk class | Examples | What applies |
|---|---|---|
| Unacceptable risk | Social scoring, manipulative systems, emotion recognition in the workplace | Prohibited |
| High risk | AI in recruitment, credit decisions or critical infrastructure | Strict obligations: risk management, documentation, human oversight |
| Transparency obligations | Chatbots, AI-generated images, videos and texts | People must be able to tell they're interacting with AI or seeing AI content |
| Minimal risk | Spam filters, text assistants for internal drafts, document analysis | No specific obligations under the AI Act |
Most applications in mid-sized companies fall into the two lower classes. But if you use AI to pre-screen job applications, for example, you quickly enter high-risk territory.
Which deadlines apply?
The AI Act applies in stages. With the so-called AI Omnibus (Regulation (EU) 2026/1744, in force since July 2026), the EU postponed some deadlines. As of November 2026:
| Since / from | What applies |
|---|---|
| 2 February 2025 | Prohibited AI practices; obligation to support employees' AI literacy |
| 2 August 2025 | Obligations for providers of general-purpose AI models |
| 2 August 2026 | Transparency obligations, such as disclosing chatbots and labeling AI content |
| 2 December 2027 | Obligations for high-risk systems under Annex III, for example in HR |
| 2 August 2028 | Obligations for high-risk AI in regulated products under Annex I |
Sources: Regulation (EU) 2024/1689, AI Omnibus overview, Future of Privacy Forum.
What does AI literacy mean for your team?
Companies using AI should take measures so their employees can use AI competently (Art. 4 AI Act). No specific certificate is required. In practice, this means: anyone working with AI should know what the tool can do, where it makes mistakes, which data may go in and when a result needs checking. A short, documented training session and an internal AI policy are a good start.
What Art. 4 requires in detail and what the 2026 AI Omnibus changed is explained in the article AI literacy training under Art. 4 AI Act.
Practical checklist
- Define which AI tools are permitted in your company and for what purposes.
- Check for each tool: where is data processed, is there a data processing agreement, are inputs used for training?
- Don't enter sensitive customer or HR data into tools that aren't approved for it.
- Assign each use case to a risk class under the AI Act.
- Label chatbots and AI-generated content that goes public.
- Train your team and keep a record of it.
- Update your privacy policy if AI processes personal data.
- Involve your data protection officer early.
An everyday example: pre-sorting customer requests with AI

Say an AI is meant to read incoming customer service emails, sort them by topic and suggest a draft reply. This is what the check looks like in practice:
- Which data? Names, email addresses and order numbers, occasionally more, for example when customers volunteer health information. So this is personal data.
- Legal basis: Handling requests usually serves the performance of a contract or a legitimate interest. This should be documented for the specific case.
- Provider: The AI service acts as a processor. A contract under Art. 28 GDPR, processing in the EU and excluding training on your data belong in the selection criteria.
- Data minimisation: Signatures, phone numbers and attachments that aren't needed for sorting are removed first.
- Human in the loop: The AI suggests, a person checks and sends. The decision stays with the team.
- AI Act: Sorting is a low-risk case. If the AI replies directly and automatically, however, customers must be able to tell they are communicating with an AI.
- Documentation: The process goes into the record of processing activities, and the privacy policy is updated.
This turns a vague concern into a manageable list you can work through before you start.
Common mistakes
- Private AI accounts for company data. Copying customer data into a private account of an AI service means no contract and no control over what happens to the data.
- Checking data protection only after the pilot. If it turns out late that the chosen provider doesn't fit, the work starts over.
- Sending all the data because it's easier. More data means more risk, but rarely better results.
- No rules for the team. Without an internal policy, everyone decides for themselves what goes into which tool.
- Publishing AI content without labelling it. The AI Act's transparency obligations have applied since August 2026.
Where can you find reliable guidance?
The German data protection authorities have jointly published guidance that is more concrete than the legal text (both in German):
- The Data Protection Conference's guidance on artificial intelligence and data protection (May 2024) is aimed at companies that select and use AI applications.
- The guidance on technical and organisational measures for AI systems (June 2025) covers what to consider when developing and operating AI systems.
How we work at uberneu
We consider data protection from the start of every AI integration: we choose models and providers to suit your data, rely on data minimization and clarify the AI Act classification before anything is built. For a legal review of your specific case, it's best to work with your data protection officer or a law firm.




