EU AI Act Transparency Rules Take Effect Today — Chatbots Must Disclose, Deepfakes Must Be Labeled

Today marks the enforcement date for Article 50 of the EU Artificial Intelligence Act — the regulation's transparency chapter, which is now binding on any provider or deployer whose AI outputs reach European users, regardless of where the company is headquartered.

The rules divide into four concrete obligations. First, interactive AI systems — chatbots, voice assistants, autonomous agents — must inform users upfront that they are talking to a machine. The exception is narrow: disclosure is only unnecessary when a reasonably informed person would already know. Deploying a customer service bot without disclosing it is AI is now, as of today, a potential regulatory offense inside the EU.

Second, providers of generative AI tools that produce synthetic audio, images, or video must embed machine-readable markings in that content and offer detection tools. The EU Code of Practice recommends two marking layers — digitally signed metadata plus an imperceptible watermark. Free-form text requires at minimum the watermark layer.

Third, any system that analyzes users' emotions or categorizes them biometrically must notify those people, clearly and at the latest at first interaction. Fourth, deepfakes and AI-generated text on matters of public interest must be explicitly labeled as artificially produced — with a narrow artistic exception where the label may appear in credits rather than inline.

The penalties are significant. Non-compliance carries up to €15 million or 3% of global annual turnover, whichever is higher. For large tech companies, the 3% of worldwide revenue figure dwarfs the flat cap, making this a real financial exposure rather than a symbolic fine.

These transparency rules are distinct from — and less demanding than — the high-risk AI system obligations found elsewhere in the AI Act. They apply to any AI system that falls into one of the four covered scenarios, whether the product is a sophisticated enterprise tool or a basic chatbot booking restaurant tables. Any EU user interaction brings a company into scope.

The European Commission confirmed that enforcement begins today, coordinated between the AI Office and national authorities. One transitional provision applies: generative AI systems already deployed before August 2 have until December 2, 2026 to add machine-readable markings to their content outputs — but the chatbot disclosure obligation applies immediately, with no grace period, as reported by the Commission's press office.

Companies that have deferred AI compliance work now face real deadline pressure. The rules are live, the enforcement machinery is in place, and the fine structure is steep enough to matter even for large organizations.