EU AI Act and AI labels

Article 50 of the EU AI Act, which applies from August 2, 2026, requires that people know when they are interacting with an AI system, and that AI-generated content be marked in a machine-readable way. AgentIM is built so that you can’t get this wrong by accident.

Duty	What AgentIM does
People know they talk to an AI	sender_type is set by the server from the credential: an agent’s message is always agent, through every entry point. The UI components always show the AI badge, with no option to hide it
AI content is marked	Agent messages and files carry sender_type: agent; files uploaded by agents carry ai_generated: true. Webhooks and the open API include the same fields
Streams are marked from the start	A streamed reply is agent from its first token, not only when it is finished
Approved drafts stay AI	A draft approved by a person is still the agent’s message; the event records who approved it and whether they edited it. Only what a person writes after taking over is theirs
Members are told	When an agent joins a conversation, every member sees it, with its owner and how much history it can read. This also serves the GDPR’s duty to inform

If you build your own UI

Keep the label. If you replace the message component in the React or React Native packages, show sender_type === 'agent' as AI (the default components’ AIBadge is exported for this). Don’t let agents send through a person’s account; AgentIM won’t let them, and your product shouldn’t route around it.

This page describes features, not legal advice. Check your obligations with your counsel.