EU AI Act Article 50: When Cold Email Actually Requires AI Disclosure

Short answer: disclosure is about interaction, not drafting. Article 50 of the EU AI Act (applicable since 2 August 2026) requires telling people when they're interacting with an AI system — it does not require labeling every email an AI helped write. The line for outbound teams falls between three scenarios:

Scenario 1 — AI-drafted, human-sent: no disclosure required

Sequences drafted with AI but reviewed and sent under human responsibility fall under the editorial-control carve-out. This covers the standard SDR workflow: AI writes, human approves, human owns the send. Keep evidence of the review step in your process docs.

Scenario 2 — Autonomous AI conversation: disclosure required

An AI agent that replies on its own — negotiating meetings, answering questions, holding the thread without human review per message — is a system the prospect is directly interacting with. Under Article 50(1) the prospect must be informed they're dealing with AI, unless it's obvious from context. If you run AI SDR reply agents into the EU, add the disclosure line.

Scenario 3 — From December 2, 2026: machine-detectability

Providers of generative AI must make outputs machine-detectable (watermarking/provenance metadata). The duty sits on OpenAI/Anthropic/Google, not on senders — but its effect reaches you: assume receiving systems can increasingly classify AI-generated text regardless of what you disclose. Deliverability strategy built on disguising automation has an expiry date.

Enforcement reality check

AI Act fines for transparency violations reach €15M or 3% of global turnover — but enforcement targets systematic deception, not individual SDR emails. The regulators most likely to reach an outbound team first remain GDPR/ePrivacy authorities (lawful basis for B2B contact), and — faster than any regulator — mailbox providers' spam filters. The technical layer is still the first gate.