Resources
Six weeks before this was written, a transparency law most of the DM automation market has not caught up with became enforceable across the EU.
Article 50 of the EU AI Act requires that anyone interacting directly with an AI system be informed of that fact — clearly, and at the start of the interaction. It applies whether the system is answering a support question, qualifying a lead, or just replying to a DM.
Article 50's transparency obligations became enforceable on 2 August 2026. That is a live compliance date, not a future one.
It was not deferred. The EU's Digital Omnibus package pushed back the Annex III "high-risk" system timeline to 2 December 2027 — but Article 50's disclosure duty was left out of that deferral. While a large part of the AI Act slipped, this part didn't.
Penalties reach €15 million or 3% of worldwide annual turnover, whichever is higher — the same tier as some of the Act's more serious obligations, not a minor administrative fine.
It doesn't ban automated replies, and the disclosure doesn't have to be apologetic or clinical. What it rules out is the version quietly sold as a real person answering — which, honestly, is what a meaningful share of the DM-automation market still does. A vendor that discloses clearly, once, at the start of a conversation, is simply complying with a law that has been enforceable since August. A vendor that doesn't is exposed to it, regardless of how good the underlying product is.
The disclosure has to happen once, at the start, in language a real person would actually read and understand — not buried in a linked policy nobody opens mid-conversation. It has to stay in place; a business shouldn't be able to switch it off to sound more human. And it should be honest in tone as well as in fact.
See how DM Handled implements this specifically on the AI disclosure doc and the compliance page.