Chatbot Transparency Obligations

What must be disclosed?

1. Disclose AI Identity

Users must clearly and unambiguously know that they are interacting with an AI system.

• "Powered by AI" or "AI Assistant" label in the chat window
• For voice: announcement "You are speaking with an AI system"
• First message: notice about AI nature

2. Label AI-Generated Content

All AI-generated outputs must be identifiable as such:

• Images: "AI-generated" watermark or metadata
• Audio: "AI-generated voice" for text-to-speech
• Video: "Deepfake" label for synthetic videos
• Text: notice in footer for articles

3. Copyright Compliance

Providers of General Purpose AI (GPAI) must document:

• Which data was used for training
• Notify copyright holders (opt-out system)
• Provide a summary of training data

4. Prevent Deception

Prohibited: Designing AI systems to deceive people.

• No human-like avatars without labeling
• No fake reviews or ratings
• No manipulative design patterns

Checklist for Businesses

Examples: Good vs. Bad

Implementation: AI Notice in Chatbot

# HTML: AI notice in chat window
<div class="chat-header">
  <span class="bot-avatar">🤖</span>
  <span class="bot-name">AI Assistant</span>
  <span class="ai-badge">Powered by AI</span>
</div>

<style>
.ai-badge {
  background: #10b981;
  color: white;
  padding: 2px 8px;
  border-radius: 4px;
  font-size: 12px;
}
</style>

# Or as the first bot message:
bot_message("""
Hello! I am an AI assistant.
My answers are generated by AI and may contain errors.

For important decisions, please verify with additional sources.
""")

Penalties for Non-Compliance

For non-compliance with transparency obligations:

• Fines up to EUR 15 million or 3% of turnover
• Warnings from data protection authorities
• Reputational damage

Sources

• EU AI Act Original Text (EUR-Lex) — Article 50
• LittleData: EU AI Act Checklist