Chatbots and AI content must be labelled: an Article 50 checklist before 2 August 2026

From 2 August 2026 Article 50 of the AI Act — the transparency obligations — applies. In practice: a chatbot must tell you that you are talking to AI; AI-generated content (audio, image, video, text) must carry a machine-readable marking; deepfakes and AI-generated text published to inform the public require disclosure. The Digital Omnibus did not move these deadlines — it deferred only the high-risk obligations and gave systems placed on the market before 2 August a transition period for machine-readable marking (until 2 December 2026). Breaching Article 50 carries fines of up to EUR 15 million or 3% of worldwide turnover (Article 99(4)).

Which paragraph applies to whom

• Article 50(1) — a chatbot on your site or in your app. The provider of a system that talks to people must disclose it is AI — unless that is obvious to a reasonably observant person. A simple “you are talking to an AI assistant” notice usually suffices.
• Article 50(2) — generating content. The provider of a generative system marks outputs as AI-generated in a machine-readable format (digitally signed metadata or an imperceptible watermark). Exception: purely assistive editing tools that do not substantially alter the input.
• Article 50(3) — emotion recognition and biometric categorisation. The deployer informs the persons exposed to the system. Note: in the workplace and in education, emotion recognition has as a rule been prohibited since February 2025 (Article 5).
• Article 50(4) — deepfakes and informative text. The deployer discloses that the content was generated or manipulated. Exceptions: evidently artistic or satirical works (limited disclosure) and text that has undergone human editorial review with editorial responsibility.

The checklist before 2 August

1. Inventory your AI–human touchpoints: website chatbots, customer-service bots, voicebots, marketing content generators.
2. Establish the role per system: provider or deployer — that decides which paragraph applies to you. The vocabulary is in the AI Act glossary.
3. Check that your chatbots introduce themselves — and that the notice is visible before the conversation starts, not buried in the terms.
4. Review published AI content: if you publish AI-generated text for informative purposes without human editorial review, you need a disclosure.
5. Plan machine-readable marking if you provide a generative tool: on 10 June 2026 the Commission published a Code of Practice with two paths — digitally signed metadata or watermarking — plus an icon set for deployers. The Code is voluntary, but it can support your evidence of compliance.
6. Assign an owner for the transparency obligations and write the rules into your AI-use policy — otherwise the arrangements die within a quarter.

Common misconceptions

“The Digital Omnibus postponed this” — no: what moved are the high-risk obligations (Annex III to 2 Dec 2027), and within Article 50 only the machine-readable marking for systems placed on the market before 2 August (until 2 Dec 2026). The full deadline overview: the Digital Omnibus and the AI Act. “We use an off-the-shelf tool, so it’s the provider’s problem” — partly: disclosing deepfakes and informative text (paragraph 4) is the deployer’s obligation, i.e. yours.

What next

How to build an agent or chatbot that meets these requirements from day one — notice, logs and oversight included — is covered in production-grade AI agents. We inventory the touchpoints and classify the roles as part of the AI Act compliance audit.