EU AI Act: chatbot and AI content labels
Article 50 of the EU AI Act sets transparency duties for AI that talks to people or makes content. A chatbot has to tell people they are talking to an AI unless that is obvious. AI-generated audio, images, video and text must carry a machine-readable mark, and deepfakes you publish must say they are artificial. These duties apply from 2 August 2026, and the Digital Omnibus on AI did not move that date.
Who has to comply
Providers, the companies that build an AI system, must make a chatbot disclose itself and mark generated content. You are a deployer if you use the system in your business: you disclose deepfakes you publish and AI text you publish to inform the public on matters of public interest, unless a person reviewed it and someone holds editorial responsibility.
Key dates
- 1 Aug 2024The AI Act enters into force across the EU.
- 2 Feb 2025Banned AI practices stop being allowed.
- 2 Aug 2025Rules for general-purpose AI models and the governance setup apply.
- 27 Jul 2026The Digital Omnibus on AI enters into force: high-risk rules move later, chatbot and labelling duties keep their date.
- 2 Aug 2026Article 50 applies: chatbots must tell people they are AI, and deepfakes must be labelled.
- 2 Dec 2026End of the four-month grace period for machine-readable marking by generative AI systems already on the market before 2 August 2026.
- 2 Dec 2027High-risk rules apply to AI used in areas such as hiring, education and biometrics.
- 2 Aug 2028High-risk rules apply to AI built into products already covered by EU safety law.
Tools that support it
Fin can show an AI Agent label, open with an AI intro message and add an AI footer to emails. The label is off by default for newer workspaces, so switch it on.
Offers a speech classifier that tells whether a clip came from ElevenLabs and says it uses the C2PA standard. Labelling a published deepfake stays your job.
Google marks content from its AI models with an invisible SynthID watermark, and Gemini can check an image for it. A visible label for your readers is up to you.
Questions
Do I need to label the chatbot on my website?
Yes, people must know they are talking to an AI unless it is obvious from the context. The law puts this design duty on the provider, but you choose how the widget looks on your site. The Commission's page says people using chatbots should be made aware they are dealing with a machine. A short first message such as "You're chatting with an AI assistant" covers it.
Does every AI image I post need a label?
No, the visible label in Article 50 covers deepfakes: images, audio or video that resemble real people, places or events and could pass as real. A clearly artificial illustration does not need one. For evidently artistic, satirical or fictional work you only disclose in a way that does not spoil the work. The tools that made the image must still add a machine-readable mark.
Do I have to say a blog post was written with AI?
Only for AI text published to inform the public on matters of public interest. Product pages, marketing copy and how-to posts are outside that duty. Even news-type text is exempt if a person reviewed or edited it and someone holds editorial responsibility. Many businesses still add a short note because readers ask.
Did the Digital Omnibus delay these rules?
No. Regulation (EU) 2026/1744 entered into force on 27 July 2026 and pushed high-risk rules to 2 December 2027 and 2 August 2028. Article 50 kept its date of 2 August 2026. Generative AI systems already on the market before that date got four months, until 2 December 2026, to add machine-readable marking.
Is there a standard way to mark AI content?
The Commission has a voluntary code of practice on marking and labelling AI-generated content. It includes a common set of icons that creators and publishers can use to show that images, audio or text are artificial. Technical marks such as C2PA Content Credentials and watermarks are how many tools add the machine-readable part. Check what your image, video or voice tool adds before you publish.