EU AI Act Article 50 hits Aug 2: label your AI content

If you run an AI chatbot on your site or ship AI-generated product copy and imagery, you have a deadline next Sunday. The EU AI Act's transparency rules — Article 50 — apply from 2 August 2026, and the European Commission refreshed its guidance on transparency of AI-generated content on July 20. This isn't the high-risk-systems regime everyone read about. Article 50 catches ordinary commercial AI use, and EU AI Act compliance here is closer to a labeling exercise than a legal project.

What actually happened

Article 50 sets four obligations, split between providers (who build and place the system) and deployers (who use it and publish the output):

• Interactive AI must announce itself. Systems that talk directly to people — support chatbots, voice agents — have to make clear the user is dealing with AI, unless it's obvious to a reasonably informed person.
• Synthetic output must be machine-readable as synthetic. Providers of generative AI must mark AI-generated audio, image, video, and text so it's detectable as artificially generated or manipulated.
• Deepfakes get a visible label. Deployers publishing AI-generated or manipulated likeness content must disclose it.
• AI-written public-interest text gets disclosed when published to inform the public and not reviewed under human editorial responsibility.

Two details that matter operationally. Generative systems already on the market before 2 August get until 2 December 2026 to meet the machine-readable marking requirement, per the Act's transitional arrangements — new systems get no runway. And the penalty tier under Article 99 tops out at €15 million or 3% of total worldwide annual turnover, whichever is higher — though for SMEs and startups, the lower of the two figures applies.

The reach is not limited to EU companies. The Act binds providers placing systems on the EU market and catches deployers and output used inside the EU, wherever you're incorporated.

Why it matters for your business

Most of this is a build task, not a legal one, and it's small if you own your stack. A disclosure line on the chat widget's first message. C2PA or equivalent provenance metadata written at generation time for AI-produced images. A flag in your CMS marking which copy came from a model. An hour or two of work if your storefront and content pipeline are yours.

It's expensive when they're not. If your chat widget is a hosted SaaS box, your product images come out of a vendor's generator, and your copy flows through a tool with no metadata hooks, then compliance is whatever your vendors decide to ship and whenever they decide to ship it. That's the real lesson here — the same one that shows up every time a rule lands. Systems you control adapt in an afternoon. Systems you rent adapt on someone else's roadmap.

If you sell into the EU, ask every AI vendor in your stack one question this week: what do you emit for Article 50, and when?

Sources: European Commission, EU Artificial Intelligence Act — Article 50.