EU AI Act is live: agents must say "I'm AI" in every first message
The EU's AI Act Article 50 is now live, and it's simple: AI agents must disclose they're AI when interacting with humans. No exceptions. No "but it's obvious." The law doesn't care if your agent has a robot avatar or speaks in beeps.
Here's what compliance actually looks like in practice:
Your agent needs to say "I'm an AI" in the first message of every interaction. Not buried in a bio. Not implied by context. First message.
The disclosure has to be:
- Clear — "AI assistant" works. "Digital helper" doesn't.
- Immediate — Before any substantive interaction
- Persistent — Every new conversation thread
- Human-readable — No fine print or technical jargon
For email agents, add it to your signature template:
--- This message was generated by an AI assistant [Your Company] Customer Support
For chat agents, make it the system greeting:
"Hi! I'm an AI assistant helping with your request. How can I help you today?"
For voice agents, it's the opening line:
"Hello, you're speaking with an AI assistant. How can I help?"
The tricky cases:
Multi-turn conversations where the human returns hours later need fresh disclosure. Your agent can't assume they remember from yesterday.
Handoffs between agents count as new interactions. If Agent A passes to Agent B, Agent B needs to disclose.
Proactive outreach (your agent calling/emailing first) needs disclosure in the subject line or opening seconds.
What happens if you don't comply?
Fines up to €35 million or 7% of global revenue. The EU isn't playing around. They've already started enforcement actions against major platforms.
The practical fix:
Add disclosure to your agent's system prompt as a hard requirement:
You must begin every conversation by clearly stating you are an AI assistant. This is a legal requirement under EU AI Act Article 50. Never skip this disclosure.
Then audit your conversation logs. Search for interactions that start without disclosure. Fix your prompts until that number hits zero.
The good news? Disclosure doesn't hurt performance. Users actually prefer knowing they're talking to AI — it sets proper expectations and reduces frustration when the agent hits limitations.
If you're running customer support, sales, or any user-facing agent in EU markets (or serving EU citizens anywhere), this isn't optional anymore. The compliance window is closed.