Skip to content
For customer service

Your support AI isn't high-risk. Article 50 still applies to it.

High-risk AI got a deadline extension to December 2027, but the Article 50 transparency rules didn't move, they've applied since August 2, 2026. If you run support chatbots, AI agents, or AI-drafted customer messages for EU users, you're already in scope. That put customer service among the first teams in any company to face the Act.
See compliance plans
01 / Article 50

Which customer service AI triggers Article 50?

Article 50 covers AI that interacts with people and AI that generates content people see. Customer service does both.

Support chatbots & virtual assistants

Any AI that converses with customers must disclose it's AI at the start of the interaction, clearly, not buried in the terms of service.

Autonomous AI support agents

Agents that resolve tickets, process refunds, or act on behalf of customers need disclosure plus logging of what they did.

AI-drafted emails and replies

AI-generated text sent to customers may need marking as AI-generated, especially when sent without meaningful human review.

Voice bots & AI phone systems

Synthetic voice interactions must disclose they're AI. Voice cloning of real agents adds deepfake obligations under Article 50(4).

02 / What to do

What you actually need to do.

The good news: this is days of work, not months. The bad news: most teams still haven't started.

  • Add a clear AI disclosure at the start of every chatbot conversation
  • Mark AI-generated content in machine-readable format (metadata, watermarking)
  • Make disclosures accessible: they must work with screen readers
  • Log AI agent actions so you can reconstruct what the system did and when
  • Update your escalation path so customers can reach a human
  • Document your approach: what's disclosed, where, and since when
03 / Not required

What's not required for limited-risk AI.

Limited risk means limited obligations. Don't let anyone sell you a conformity assessment you don't need.

  • No conformity assessment, that's only for high-risk systems
  • No EU database registration
  • No quality management system requirement
  • No mandatory third-party audit
  • Obvious AI use may be exempt, but 'obvious' is judged from the user's perspective, not yours

One caveat: if your support AI makes decisions that affect access to services (auto-denying refunds, closing accounts, restricting essential services), it may cross into high-risk territory. Run the classifier to check.

04 / Vendors

Using Intercom, Zendesk, or another platform?

If you deploy a third-party AI support tool, you're a deployer with your own obligations. The platform may provide the disclosure mechanisms, but you're responsible for making sure they're enabled, visible, and appropriate for your use. Check what your vendor provides, document the gap, and close it.

Read the deployer obligations guide
Further reading
5 min read

Your Chatbot Needs an AI Disclosure. Here's Exactly What to Write.

7 min read

EU AI Act and AI Agents: When Autonomous Systems Trigger Compliance Obligations

6 min read

Article 50 Transparency Obligations: Your Compliance Checklist

Article 50 is already in force.

Adding disclosures takes an afternoon. Knowing exactly which of your AI tools need them takes a few questions with the classifier. Start there.

Classify your AI system