Since August 2, 2026, the European AI Office and national authorities can impose penalties. Any system that talks to people must identify itself as artificial, AI-generated content needs a visible label and a machine-readable mark, and general-purpose model providers must publish a detailed summary of the content they trained on.
On August 2, 2026 the section of the EU Artificial Intelligence Act that gives real enforcement powers to the European AI Office and to each member state's designated authorities came into force. Until then the regulation existed on paper; from that date, it exists with fines behind it.
The transparency obligations, specifically:
The penalty regime: up to €15 million or 3% of worldwide annual turnover — whichever is higher. More than 180 signatories have joined the voluntary Code of Practice, which works as a fast lane for demonstrating compliance.
What comes next: December 2, 2026 brings the ban on AI-generated non-consensual intimate imagery and child sexual abuse material; December 2, 2027 brings the high-risk system obligations; and August 2, 2028 closes the implementation calendar.
Why it matters: this is the first time an economic bloc of 450 million people can fine an AI lab for failing to explain what it trained on. Everything built for the European market — including products from Anthropic, OpenAI and Google — will from now on be designed with this text in view.
Read the Commission's official release →Source: European Commission · Help Net Security
Go to the original source →