The EU AI Act's big deadline moved. The chatbot rules didn't.

What didn't move are the Article 50 transparency obligations. From 2 August, people in the EU interacting with an AI system need to be told they're dealing with AI, and generated content is expected to be marked as such. The one concession is a four-month grace period, to 2 December 2026, for the watermarking requirement on systems already on the market.

The omnibus also added a ban on AI "nudifier" tools and AI-generated child sexual abuse material, with a transition to 2 December 2026, and moved the deadline for national AI regulatory sandboxes to August 2027.

Who this actually affects

More small businesses than you'd think. If you have customers or users in the EU and any of these, the transparency rules are your bit of the Act:

• a chat widget or support bot that answers with AI
• an AI writing, summarising or image feature inside your product
• AI-generated images, audio or video that you publish or hand to clients

Agencies have an extra angle. If you build chatbots or content tools for clients, the disclosure has to exist in what you ship, and the client will expect you to know that. "We hadn't heard" isn't a great line in a renewal meeting.

What to do this month

This is a design job more than a legal one, and it's quicker to do now than after a complaint:

1. List every place a person meets AI in your product, your site or your client work: chat, email replies, generated images, voice.
2. Add a plain disclosure where people actually see it, not buried in the terms. Something like: "You're chatting with an AI assistant. Ask for a person at any time."
3. Decide how you'll mark generated images and media: a caption, a label, metadata, or all three. Note where you'll need watermarking by 2 December.
4. Add a line to your client proposals and handover documents covering AI disclosure, so it's agreed rather than assumed.
5. If anything you build touches hiring, credit or education decisions, book a conversation with a lawyer well before December 2027. That's the high-risk list, and it carries much heavier duties.

This is a summary, not legal advice. But for most founders and agencies reading this, the honest version is short: tell people when they're talking to AI, label what AI made, and write down that you've done both.