Article 50 · Agents
Does an AI agent need to disclose it is AI?
Not legal advice. ActBrief prepares a preliminary technical readiness brief for legal review and enterprise diligence — not a conformity assessment, certification, or guarantee of compliance.
Short answer
If a natural person in scope interacts with your agent or receives communications that are AI-generated / agent-sent in a way Article 50 covers, plan for disclosure and evidence. Autonomy does not remove transparency — it usually adds oversight and logging questions. Confirm with counsel.
1. Map the surfaces (not the buzzword)
- In-product agent chat — same Art. 50(1) interaction pattern as a chatbot. Wording · Placement
- Outbound email / Slack / LinkedIn — disclose in the message itself when the recipient is interacting with / receiving AI output
- Tool actions with no human in the thread — still document role, intended purpose, and whether any person-facing output exists
- Generated reports / docs — may raise Art. 50(2)/(4) questions depending on content type and who publishes
2. Starting wording (agents)
In-product agent
You are working with an AI agent. It can take actions in [systems]. Outputs and actions may be wrong. Confirm high-impact steps or contact [human path].
Outbound message
This message was generated by an AI system operated by [Company]. It may be incomplete or wrong. Reply [HUMAN] or contact [email] for a person.
More copy: AI disclosure notice template · disclosure examples.
3. Provider vs deployer for agents
Shipping an agent that customers use often means you place an AI system on the market (provider of that system) even if the model comes from OpenAI/Anthropic. Role map: OpenAI-powered SaaS · provider vs deployer vs GPAI.
4. Beyond disclosure
Agents that affect people (recruiting, credit, essential services) may also raise Annex III / Art. 6(3) questions — separate from Article 50. Oversight and logging matter more as autonomy rises. Broader checklist: AI Act for AI agents · SaaS compliance checklist.
5. Evidence to keep
- Screenshot or email sample of live disclosure
- List of agent capabilities / tools
- Product version + owner + approval date
- Art. 50 paragraph mapped per surface
Evidence log template · Article 50 Evidence Pack.
FAQ
Does an AI agent need to disclose it is AI?
Often yes when a natural person interacts with the agent or receives agent-generated communications and Article 50 applies. Obligations have applied since 2 August 2026. Confirm paragraph, actor, and exceptions with counsel.
What if the agent only acts inside our SaaS UI?
In-product chat or “do this for me” assistants are usually interaction surfaces — treat them like chatbots for Art. 50(1) mapping. See chatbot disclosure and placement guides.
What about emails or Slack messages the agent sends?
Outbound agent messages are a common gap. Prefer an explicit AI signature in the message body or signature block, plus a human contact path. Do not rely on a marketing-site footer.
Is the model vendor responsible for agent disclosure?
Usually not for your product surface. Calling OpenAI/Anthropic does not make you “only a deployer” of their model for every duty. Map roles per feature.