Transparency Obligations

Transparency is a cornerstone of the EU AI Act. Regardless of risk level, certain information and labeling obligations apply — particularly relevant for companies creating AI-generated content or deploying chatbots.

General Transparency Obligations (Art. 50)

Chatbots and Conversational AI

When users interact with an AI system, they must be informed:

• ✅ Clear notice: "You are communicating with an AI"
• ✅ Before or at the beginning of the interaction
• ✅ In understandable language
• ❌ Not hidden in terms of service or footer

Exception: When obvious from context (e.g., AI voice assistant named "AI Assistant").

AI-Generated Content

Texts, images, audio, and video generated by AI must be machine-readably labeled:

• Texts: Metadata or watermarks indicating AI origin
• Images: C2PA standard or comparable technical marking
• Audio: Audible or machine-readable labeling
• Video/Deepfakes: Clear labeling as AI-generated

Emotion Recognition Systems

Where permitted (not in workplaces/educational institutions):

• Users must be informed in advance
• Purpose of recognition must be stated
• Data protection consent required

GPAI Rules (General Purpose AI)

Since August 2025, special rules apply to Foundation Models and General Purpose AI:

Obligations for All GPAI Providers

• Create and maintain technical documentation
• Provide information for downstream providers
• Comply with EU copyright directive (opt-out mechanism)
• Publish summary of training data

Additional Obligations for GPAI with Systemic Risk

GPAI models with systemic risk (training above 10^25 FLOP or equivalent):

• Model evaluation: Standardized testing and red-teaming
• Risk assessment and mitigation: Documented measures against systemic risks
• Incident reporting: Report serious incidents to the EU AI Authority
• Cybersecurity: Adequate protection level for model and infrastructure

Information Obligations for Deployers

As a deployer of an AI system, you must:

• Inform employees: When AI is used in the workplace
• Inform affected persons: When AI decisions affect individuals
• Involve works council: In Germany per BetrVG when introducing AI
• Data protection impact assessment: Update when AI processes personal data

Implementation in Practice

Practical Tip: Implement the chatbot notice today. It's the easiest compliance measure and shows your customers that you take transparency seriously.