Article 50 transparency obligations are live as of August 2, 2026.
Deployers of AI systems that interact directly with people must now inform those people they are interacting with AI, and must disclose when content has been AI-generated or manipulated. These are not optional. Enforcement of the EU AI Act began the same day, and non-compliance carries fines of up to EUR 15 million or 3 percent of worldwide annual turnover, whichever is higher.
What this requires in practice.
The obligation is not just to tell people. It is to be able to prove you told them. A regulator will not ask whether your policy says you disclose AI involvement. They will ask you to produce the record: which system, which interaction, on what date, with what disclosure language, delivered to whom. That is an evidence problem, and it must be solved at the moment the AI touches the user, not after an inquiry lands.
Translate the model to the regulator.
Norma sits in the API path and observes each decision as it happens. For every AI output it captures the reasoning, the confidence, the risk flags, and the disclosure record, then packages them as regulator-ready evidence in plain language a compliance officer can defend, without needing a machine-learning team.
Produce the evidence a regulator will ask for at norma.northgale.io.