What this screen covers
Article 50 of Regulation (EU) 2024/1689 is a transparency rule, not a general AI Act readiness test and not a GPAI role test. This page asks bounded questions about the organisation's role and the kind of system, then lists which disclosure duties may be in play.
Application date
Article 50 applies from 2 August 2026. That date is informational. It is not a current release from the duties, and this screen does not calculate a countdown score.
Provider and deployer
A provider designs or develops the system. A deployer uses it. The same organisation can be both. Provider marking of synthetic output is a different duty from a deployer telling a person what they are seeing or hearing.
Chatbots and direct interaction
Where an AI system is intended to interact directly with a person, people are to be told they are interacting with AI, unless that fact is obvious to a reasonably informed, observant, and circumspect person in the context.
Machine-readable marking
A provider of a system that generates synthetic audio, image, video, or text is to mark outputs in a machine-readable way and make them detectable as AI-generated or manipulated, as far as technically feasible. Limited assistive editing that does not substantially change the input is a review note, not an automatic release.
Emotion recognition
A deployer of an emotion-recognition system informs the natural persons exposed to it, subject to the statutory limits. This screen records whether that notice exists. It does not store faces, voice, or other biometric samples.
Biometric categorisation
The same exposed-person notice question applies to biometric categorisation. Saying that a file was machine-marked does not answer whether the person was told.
Deepfake disclosure
A deployer of a system that generates or manipulates a deepfake discloses that the content was artificially generated or manipulated. This tool does not confirm that a file is a deepfake. It screens the duty if you have already classed the content that way.
Public-interest text
A deployer who publishes AI-generated or manipulated text to inform the public on a matter of public interest discloses that fact. The human-review path is separate and is described below.
Human review
The public-interest text duty has a narrow path where the text had human review or editorial control and a natural or legal person holds editorial responsibility. Someone glancing at a draft is not that record. Unknown responsibility stays a review item, and the path does not cancel a separate marking duty.
Artistic and creative works
For a work that is evidently artistic, creative, satirical, fictional, or analogous, the deepfake disclosure can be adapted so it does not hamper display or enjoyment. The disclosure still has to exist.
Timing and accessibility
Information under Article 50(1) to (4) is to be clear and distinguishable, at the latest at the first interaction or exposure, and it has to meet the accessibility requirements that apply.
How the screen reads common cases
- A customer-service chatbot whose AI nature is not obvious can require an interaction disclosure.
- A provider that generates synthetic images without machine-readable marking can have both the marking duty and a marking gap.
- A deployer that publishes a deepfake video can require a deepfake disclosure.
- A clearly fictional creative work can allow an adapted disclosure. It does not remove the disclosure.
- Public-interest text with human editorial review and a named editorial responsibility can show that the editorial path may apply.
- The same text without human review can require the public-interest disclosure.
Limits
Answers stay inside the enums on the form. There is no score, percentage, or probability. A result is not a certificate, a safe-harbour finding, or a confirmation that content is a deepfake. Commission guidance and the transparency code are review triggers for a manual ruleset update.