EU AI Act · Article 50 · In force
Article 50 is not waiting for the high-risk deferral. From 2 August 2026, transparency obligations — chatbot disclosure, synthetic-content marking, deepfake labelling — are enforceable. If your product or marketing uses AI, read this.
The duties
AI systems intended to interact directly with natural persons must be designed so that the person is informed they are interacting with an AI — unless this is obvious from context.
In practice: A customer-service chatbot must disclose it is AI. The disclosure must be clear and in place at the first interaction.
Systems generating synthetic audio, image, video, or text must mark the output in a machine-readable format as artificially generated or manipulated.
In practice: AI-generated product images, synthetic voiceovers, and generated text must carry a detectable marker (watermark, metadata, or equivalent technical solution).
Deployers of emotion-recognition or biometric categorisation systems must inform the exposed persons, except where the law permits it.
In practice: If you use AI that infers emotion or categorises people by biometric traits, the people affected must be told.
Deployers of deepfakes must disclose that the content is artificially generated or manipulated.
In practice: A synthetic spokesperson video or an AI-generated avatar must be disclosed as artificial.
Text published to inform the public on matters of public interest, generated by AI, must be disclosed as artificially generated — unless there is human editorial review and responsibility.
In practice: AI-drafted news or public statements need a disclosure, unless a human editor takes responsibility for the content.
The hard part
The synthetic-content duty is not a footer disclaimer. It requires the output to carry a detectable marker — a watermark, metadata tag, or equivalent technical solution — so a downstream system or regulator can identify it as AI-generated. That means it has to be built into your generation pipeline, not added as a policy sentence.
The same is true of chatbot disclosure: it must be present, clear, and at the first interaction. If your chatbot sits behind a SaaS widget, you are still the deployer responsible for the disclosure.
A €500 readiness audit maps which Article 50 duties apply to your systems and how to evidence them. Ten business days, and you keep the documentation.
FAQ
From 2 August 2026, and they are not deferred by the Digital Omnibus. They apply to providers (chatbot disclosure, synthetic-content marking) and deployers (emotion-recognition disclosure, deepfake labelling, and AI-generated public-interest text).
Yes. AI systems intended to interact directly with natural persons must be designed so the person is informed they are interacting with an AI, unless this is obvious from the context. The disclosure must be present, clear, and in place at the first interaction — even when the chatbot sits behind a SaaS widget, you remain the deployer responsible for it.
AI-generated audio, image, video, or text must carry a detectable marker — a watermark, metadata tag, or equivalent technical solution — so a downstream system or regulator can identify it as artificial. It must be built into the generation pipeline, not added as a policy sentence or footer disclaimer.
Text published to inform the public on matters of public interest that is generated by AI must be disclosed as artificially generated, unless there is human editorial review and responsibility. Deepfakes must also be disclosed as artificial.