EU AI Act transparency rules now apply to your app's AI chat features
Since August 2026, EU users must be told when they talk to AI, and existing generators must mark output by Dec 2, 2026. What the Digital Omnibus delayed.
The EU AI Act became applicable on August 2, 2026, with some exceptions. Shortly before that, on July 27, the Digital Omnibus on AI entered into force and moved several deadlines, mostly for high-risk systems. If your app ships a chatbot, an AI assistant or a feature that generates images, audio or text for EU users, the transparency rules in Article 50 are the part that applies to you now. What follows is an engineer's reading of the sources on what applies, what moved and what to change, and it is not legal advice.
What changed
The AI Act applies, with exceptions
The European Commission's summary is that the AI Act "became applicable on 2 August 2026, with some exceptions", and that "the transparency rules of the AI Act will come into effect in August 2026". The exceptions matter, because the Digital Omnibus moved the high-risk obligations to later dates.
The Digital Omnibus moved the high-risk dates
The Digital Omnibus on AI is Regulation (EU) 2026/1744. It was adopted on July 8, 2026, published in the Official Journal on July 24, 2026, and entered into force on July 27, 2026.
Its main effect is time. High-risk obligations for Annex III systems, which cover areas such as employment, credit and education, move from August 2, 2026 to December 2, 2027. Obligations for AI embedded in Annex I products move to August 2, 2028.
The Omnibus also adds new prohibitions on AI that generates non-consensual intimate imagery of identifiable people and child sexual abuse material. It softens the AI-literacy duty to "support the development of AI literacy". And it requires the Commission to create simplified technical documentation forms for SMEs and small mid-caps.
Article 50 transparency
Two parts of Article 50 matter for typical apps. Article 50(1) requires that AI systems interacting with people inform users that they are dealing with AI, "unless this is obvious" to a reasonably well-informed person.
Article 50(2) requires that synthetic audio, image, video or text be marked "in a machine-readable format and detectable as artificially generated or manipulated". Systems already on the market before August 2, 2026 have until December 2, 2026. That extra time applies only to those existing systems.
The fines are set in Article 99. Breaking Article 50 can cost "up to EUR 15 000 000 or ... up to 3 % of its total worldwide annual turnover", whichever is higher.
What it means for your app
If you ship an in-app chatbot, AI assistant or support bot to EU users, you need a clear "you are talking to AI" disclosure now. The "unless this is obvious" exception is a judgment call, and I would not build a product decision on it. A short label in the chat header and in the first message costs very little.
If a feature generates images, audio or text, you need a plan for machine-readable marking, such as watermarks or C2PA. For features that were live before August 2, 2026, the deadline is December 2, 2026. New generation features should ship with marking from day one.
Text is in scope. Article 50(2) lists synthetic text next to audio, images and video, so check every place where your app produces text with a model, including summaries, suggested replies and generated product descriptions.
The new prohibitions affect any app that lets users generate images of people. The Omnibus prohibits AI that generates non-consensual intimate imagery of identifiable people or child sexual abuse material, so filters on prompts and outputs now belong in your compliance work as well as your trust and safety work.
Most indie and agency apps are not high-risk. If you build HR, credit or education products, though, the Omnibus gave you a reprieve until December 2027, and you should use that time to prepare. Small studios also get some relief on paperwork, because the Commission must create simplified technical documentation forms for SMEs and small mid-caps.
Compliance checklist
- List every feature in your apps where a user interacts with an AI system or receives AI-generated audio, images, video or text.
- Add a visible AI disclosure to every chat, assistant or support bot that EU users can reach.
- For each generation feature, record whether it was on the market before August 2, 2026, because that decides whether the December 2, 2026 date applies.
- Choose a machine-readable marking approach, such as watermarks or C2PA, and check whether your model provider already supports it.
- Apply marking in your server pipeline, so that every generated asset is marked before it reaches the client.
- Block prompts and outputs that would produce non-consensual intimate imagery of identifiable people or child sexual abuse material, and log the rejections.
- If any client works in employment, credit or education, flag the project for a high-risk review well before December 2, 2027.
- Write down these decisions in a short internal note, and get legal review for anything close to the high-risk categories.
My take
The Omnibus made the AI Act easier for most app developers, and the transparency rules are not hard to meet. Disclosure is a UI change. Marking is a pipeline change that you can make once, on the server, for every generation feature.
The risk for small teams is reading "high-risk delayed" as "AI Act delayed". The disclosure duty applies now, and the December 2, 2026 deadline for marking existing features is close. Given the Article 99 fines, a label and a marking step are the cheaper option.
I also expect the details of marking to keep moving as tools and guidance mature, so build it as a replaceable step in your pipeline and avoid hard-coding a single vendor's format into your clients.