Don’t forget to disclose your chatbot — Article 50 lands 2 August and it was not deferred

From 2 August 2026, EU AI Act Article 50 transparency obligations apply — users must be told when they are interacting with an AI system or looking at AI-generated media. The digital omnibus deferred the high-risk regime; it did not defer this. If your chatbot stays silent, you are the one exposed.

Who this is for

• Anyone running a customer-facing chatbot, voice assistant or support agent in the EU market.
• Product and engineering leads who ship AI features and assumed the omnibus bought them time.
• Compliance and risk owners who need a defensible answer on what applies on 2 August.
• Marketing and content teams publishing AI-generated or AI-edited images, audio or video.

What’s happening

Article 50 of the EU AI Act sets the transparency duties: a system that interacts with people must make clear they are dealing with an AI unless it is obvious; synthetic audio, image and video content must be marked as artificially generated or manipulated; and deployers of emotion-recognition or biometric-categorisation systems must inform the people exposed to them. Those obligations become applicable on 2 August 2026.

The confusion comes from the digital omnibus. It pushed the high-risk obligations back — Annex III high-risk to 2 December 2027, Annex I embedded high-risk to 2 August 2028. Article 50 was not part of that reset. It sits on the original date and it is now binding law. Read the omnibus as a deferral of the heavy conformity work, not a general amnesty.

Why it bites

Transparency looks like the easy tier, and that is exactly why it gets missed. The duty falls on the deployer — the organisation putting the system in front of users — not only on whoever built the model. A generic “powered by AI” footer is not the same as telling a user, at the point of interaction, that the thing answering them is a machine. And AI-media labelling is a duty most content teams have never been briefed on, so the exposure is spread across functions that do not think of themselves as running an AI system at all.

The one thing to do before 2 August

Inventory every point where a user meets your AI and confirm each one carries a clear disclosure. Walk the actual journeys — the web chat, the in-app assistant, the phone line, the auto-generated images on your campaigns — and write down, for each, the exact words shown and when they appear. If any surface has no disclosure, or hides it behind a settings menu, fix that one first. I set out what a disclosure that holds up actually looks like in this piece on building Article 50 disclosure.

If you want a second pair of eyes on that inventory before the date, a technology control review covers exactly this: mapping the interaction points, checking the disclosures against the obligation, and leaving you with evidence you can show a regulator.

Engagements are fixed-scope and priced up front, so you know the cost before you commit.

Transparency is the part of the AI Act you can get right with a checklist rather than a programme — but only if you start before 2 August 2026. That is the date. It did not move.