AI transparency · 8 minute read
EU AI Act Article 50: what must be disclosed from August 2026?
Article 50 transparency obligations started applying on 2 August 2026. The rules do not say that every use of AI needs the same label: the duty depends on whether you provide an AI system, deploy one, or publish particular kinds of generated or manipulated content.
First identify your role and the output
Providers of systems intended to interact directly with people must generally design them so people are told they are interacting with AI, unless that is obvious to a reasonably informed and attentive user. Providers of systems that generate synthetic audio, images, video or text have a separate duty to make outputs detectable and mark them in a machine-readable format.
Deployers have different disclosure duties for emotion-recognition or biometric-categorisation systems, deepfakes, and certain AI-generated or manipulated text published to inform the public about matters of public interest. A single generic footer is therefore not a reliable answer for every product or publication.
Visible labels and machine-readable marking are different layers
A person-facing disclosure explains that an interaction or item was generated or manipulated with AI. Machine-readable marking is intended to make synthetic output detectable by technical systems. Article 50 can require one or both layers depending on the situation.
The Commission says disclosures should be clear and distinguishable and provided no later than the first interaction or exposure. The exact implementation still needs to fit the medium, accessibility needs and the provider or deployer role.
- State what was generated or altered
- Place the notice where a person will encounter it in time
- Keep the wording understandable and accessible
- Do not treat visible text as a substitute for a provider’s technical marking duty
Editorial review can matter for public-interest text
The Article 50 rule for AI-generated or manipulated text published to inform the public about matters of public interest includes an exception where the content has undergone human review or editorial control and a person or organisation holds editorial responsibility for publication.
That exception is specific; it is not a general statement that human review removes every AI transparency obligation. Deepfake rules and provider-side marking duties should be assessed separately.
Use a documented review instead of guessing
Record the system, the content type, who generated or deployed it, whether people interact with it, whether it depicts real people or events, the audience and the review process. Then map those facts to the relevant paragraph of Article 50 and the Commission guidance.
The 4g3n7 checker and disclosure generator can organise that first review, but they are not a compliance certificate or legal opinion. Consequential or uncertain cases need qualified advice.