AI Act Since August 2026: What Chatbots and AI Content Must Disclose

Article 50 of the European Union's AI Act has applied since August 2, 2026. Shortly before that date, the European Commission published guidance and updated its FAQ. Those documents provide the Commission's current but non-binding interpretation. Neither the law nor the guidance creates one universal visible AI label for every generative use. The applicable duty depends on the system's function, the content, and the actor's legal role.

A provider of an AI system that interacts directly with people generally must inform them that they are communicating with AI. Under Article 50(5), the notice is due no later than the first contact; the Commission places it at the beginning of the first interaction. The exception applies only when the AI interaction is obvious to a reasonably well-informed, observant, and circumspect person from the circumstances and context. Under the Commission's interpretation, a notice that appears only after several exchanges or solely in terms and documentation is not enough. The rule requires direct interaction. Background processing or machine-to-machine communication without direct human contact does not fall under paragraph 1 merely because AI is involved. An organization that places the system on the market or puts it into service under its own name or trademark may itself be the provider. The underlying model vendor does not determine that role by itself.

Machine-Readable Marking

Article 50(2) covers systems that generate synthetic text, image, audio, or video. Their outputs must be marked in a machine-readable format and made detectable as artificially generated or manipulated. To the extent technically feasible, the solution must be effective, interoperable, robust, and reliable. A visible caption may inform a person, but it does not replace the technical layer. The statute excludes standard editing that does not substantially alter the input data or its meaning; the Commission guidance addresses additional edge cases. For systems placed on the market before August 2, 2026, a limited transition applies: the paragraph 2 duty must be met by December 2, 2026. The transition does not generally postpone the other transparency obligations.

Disclosure by Deployers

A deployer that uses an AI system to generate or manipulate a deepfake must visibly disclose its artificial origin. The disclosure must be clear and distinguishable no later than the first exposure. An invisible provenance mark is not enough for the audience. The AI Act defines a deepfake as AI-generated or manipulated image, audio, or video that resembles existing people, objects, places, entities, or events and falsely appears authentic or truthful. The Commission interprets existing broadly enough to include plausible real-world subjects in the circumstances described by its guidance.

Another deployer duty covers AI-generated or manipulated text published to inform the public about matters of public interest. The law provides an exception when the text has undergone human review or editorial control and a natural or legal person holds editorial responsibility. Both conditions must be met. The Commission says spelling or grammar correction alone is insufficient; substantive judgment and authority to revise or reject publication are required. That exception addresses the deployer's visible disclosure of the text. It does not cancel the provider's separate technical marking duty. Conversely, a machine-readable mark does not replace visible deepfake disclosure. A voluntary code of practice or Commission-provided symbols also does not change the statutory assignment of roles.

Implementation therefore requires more than an inventory of AI products. Teams can test the paths through which output actually reaches people or downstream systems. A notice may disappear in an embedded interface, a mark may be lost during conversion or re-upload, and an internal draft may later become public-interest content. Download, format conversion, and publication are useful test points for that approach. The duty in any particular case still depends on the system, role, content, and use context.

Note: This assessment is not a substitute for a review of the specific case.