AI Regulation Tracker  /  European Union

EU Commission Issues Draft Guidelines on AI Act Article 50 Transparency

On May 8, 2026 the European Commission published draft guidelines explaining how the transparency duties in Article 50 of the AI Act actually work: telling people when they are talking to a chatbot, marking AI-generated content so machines can detect it, and labelling deepfakes. The guidelines are non-binding, but the duties they interpret are binding and apply from August 2, 2026. If your company runs a chatbot, generates synthetic media, or ships genAI features to users in the EU, this is the document that tells you what compliance looks like.

By Anthony Guerriero  ·  The Leveraged Years Regulatory Research  ·  Published July 20, 2026  ·  Last updated July 20, 2026

If you build a chatbot, a genAI writing tool, or anything that produces synthetic images, audio, or video, and any of your users sit in the European Union, the EU just handed you the answer key. On May 8, 2026 the European Commission published draft guidelines on how Article 50 of the AI Act works. Article 50 is the transparency chapter, the part that says people should know when they are dealing with a machine and when a piece of media was made or altered by one. The duties themselves were already written into the AI Act. What was missing was a clear read on what they require in practice, and that is what these guidelines fill in.

I want to be straight about status. This is a draft, not a final text. The Commission ran a public consultation that closed on June 3, 2026, and a final version is expected before the duties bite. The guidelines are also non-binding by design. But that does not make them optional reading. The Commission describes them, in the document itself, as "practical guidance to assist competent authorities, as well as providers and deployers of AI systems, in ensuring compliance with the transparency obligations under Article 50 AI Act in a consistent, effective and uniform manner." When a regulator tells you how it reads a law, you plan around that reading.

How this differs from the Code of Practice

There are two EU documents circling Article 50 right now, and people mix them up. The Code of Practice on transparency of AI-generated content is the technical one. It covers how you actually mark and label content so a machine can detect it. These guidelines are the legal one. They explain who is caught, what each duty means, and where the edges are. The Commission puts the split plainly: it prepared the guidelines in parallel to the Code, and the guidelines "clarify the scope of the legal obligations and addressing aspects not covered by the code." Read the Code for the plumbing. Read the guidelines for the scope.

The chatbot disclosure duty

Article 50(1) is the one most US companies will touch first. If your AI system is meant to interact directly with people, you have to make sure those people know they are dealing with an AI. The guidelines are firm about what does not count. A line buried in your terms and conditions does not satisfy it. A metadata tag or watermark on its own does not satisfy it. A vague reference to an "assistant" does not satisfy it. The disclosure has to be perceivable inside the interaction itself, through means like clearly visible plain-language notices, audio cues, or persistent visual indicators. The bar is higher when children or other vulnerable users are in the audience. The guidelines also stretch this to agentic AI that acts on a user's behalf, which is worth noting if you are shipping agents this year.

Marking and deepfake labelling

Article 50(2) puts a duty on providers of generative systems to mark their outputs so that content is detectable as AI-generated or manipulated, in a machine-readable format. That is the piece the Code of Practice supports on the technical side, and it is also the piece a proposed AI Omnibus package would give a short deferral to, from August to December 2026. That deferral is a proposal, not law, so plan for August unless and until it is formally adopted.

Article 50(4) is the deepfake and public-interest-text duty, and it lands on deployers. Two points stand out. First, the labelling duty applies even without any intent to deceive. If you generate or edit realistic content that depicts people, places, or objects that could exist in reality, you label it, full stop. Clearly unrealistic material, the flying-elephant category, is out of scope. Second, there is a narrow carve-out for AI-assisted text published to inform the public. To use it, the content needs genuine substantive review by a qualified human and clearly attributable editorial responsibility, with a name and contact that the public can actually find. This is a real gate, not a loophole.

Why this reaches US companies

Do not read Article 50 as an internal EU matter. The AI Act follows the market, not the mailing address. The transparency duties apply to providers and deployers whose systems are placed on the EU market or whose output is used in the EU. A US SaaS company running a customer-service chatbot that European users log into is a deployer, and often a provider too. A US genAI vendor whose image or text generator produces content consumed in the EU is in the marking chain. The obligations can also stack. The guidelines confirm that more than one Article 50 duty can apply to the same system at once, so a single product might owe both interaction disclosure and content marking.

One more thing the guidelines flag that is easy to miss. Meeting the AI Act alone does not clear you under every other regime. Consumer protection, platform rules, and personality and IP rights can each carry their own labelling and disclosure demands. Article 50 is a floor, not a ceiling.

What to do before August 2

Treat this as a short punch list you can hand to product and legal.

• Map where your AI touches EU users. Chatbots, assistants, agents, and any generative feature. If EU users can reach it, it is in scope.
• Move disclosure into the interaction. Get the "you are talking to an AI" signal out of the fine print and into the interface, in plain language, visible or audible at the moment of contact.
• Turn on machine-readable marking for AI-generated and edited outputs, and follow the Code of Practice for the technical how.
• Set a deepfake labelling rule that triggers on realistic synthetic media regardless of intent, and document your editorial-review process if you plan to rely on the news carve-out.
• Check the stacking and the other regimes. Confirm which Article 50 duties apply together, and whether consumer or IP rules add more.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. These guidelines were in draft form when published here; confirm the final text and how Article 50 applies to your products with qualified EU counsel.