Transparency Obligations
🖱️ Interaktiv: Ziehen zum Drehen · Lektion 4
Learning objectives
After this lesson you will be able to:
- Correctly label AI-generated content
- Implement the requirements for human oversight
- Create an AI inventory for your organization
- Meet the documentation duties as a deployer of AI systems
Labeling obligations
The EU AI Act obliges deployers of AI systems to be transparent toward:
- Employees: the use of AI in the workplace must be communicated
- Customers: interaction with an AI system (chatbot) must be recognizable
- The public: AI-generated content (text, images, video) must be labeled
💡 Practical labeling
• Chatbot: "I am an AI assistant" at the start of the conversation • AI-generated text: note "This text was created with AI assistance" • AI images: watermark or metadata "AI-generated" • AI decisions: "This decision was made automatically"
Labeling AI-generated content
The labeling must:
- Be clear and easily visible (not in the fine print)
- Occur at the time of interaction
- Be understandable to the recipient (no jargon)
- Not be misleading
Human oversight
Human oversight means that a person retains control over AI systems. It is legally required for high-risk systems, but strongly recommended for low-risk systems too.
Forms of human oversight
- Human-in-the-loop (HITL): the human must confirm every AI decision
- Human-on-the-loop (HOTL): the human monitors the system and intervenes as needed
- Human-in-command (HIC): the human makes strategic decisions about AI use
📝 Schnellprüfung
What does 'human-on-the-loop' mean?
AI inventory — the AI register
An AI inventory is the systematic recording of all AI systems in your organization. The EU Commission recommends it as the first step toward meeting Art. 4.
What belongs in the AI inventory?
| Field | Description | Example |
|---|---|---|
| Name | Designation of the AI tool | ChatGPT Enterprise |
| Type | LLM, image generation, etc. | Language model |
| Vendor | Provider of the system | OpenAI |
| Risk class | Under the EU AI Act | Minimal risk |
| Use case | What it is used for | Text drafts |
| Responsible party | Who is accountable | Department head |
🏢 Praxis-Szenario: Building an AI inventory
A marketing agency with 50 employees uses: ChatGPT for texts, Midjourney for images, Copilot for code, an AI-based translator, and an AI tool for customer analytics. One employee also uses another AI tool on their own (shadow AI). How do you proceed?
Documentation duties
As a deployer of an AI system you must document:
- Which AI systems are in use?
- For what purposes are they used?
- What data is used?
- How is quality ensured?
- What human oversight measures exist?
- What training has been conducted?
The documentation must be presentable in the event of an inspection by the supervisory authority.
💡 Tip: documentation as a process
Documentation should not be treated as a one-time task but as a continuous process. Keep a living document that you update regularly — not a PDF that gets outdated and forgotten.
✅ Wichtige Erkenntnisse
Haken setzen, um deinen Lernfortschritt zu markieren:
→ Go deeper: Data protection & GDPR — data protection impact assessments and DPAs · AI tools at work — shadow AI and practical implementation
Weiterführende Inhalte
Relevante Rechtsgrundlagen: Art. 13 · Art. 14 · Art. 50 · Art. 52