EU passes historic AI Act forcing sweeping controls on frontier models

Brussels, 6 June — In a landmark session that drew observers from Washington to Beijing, the European Parliament approved the AI Act by a margin few analysts had predicted. The legislation, seven years in drafting, creates a four-tier risk framework that categorises AI applications from minimal to unacceptable risk, with the latter category — including real-time biometric surveillance in public spaces — banned outright.

The vote was greeted with applause in the chamber and immediate concern from Silicon Valley. Shares in Alphabet, Meta and OpenAI's nearest public proxy all fell between 2 and 4 percent in after-hours trading. Industry groups pledged to challenge provisions around general-purpose AI models, which now face mandatory capability evaluations before deployment in EU markets.

Under the new rules, frontier model developers must submit to independent audits, publish detailed training data summaries and maintain incident-response registers accessible to national authorities within 72 hours of a serious breach. Penalties for violations scale up to 35 million euros or 7 percent of global annual turnover — whichever is higher.

Prime Minister Mette Frederiksen of Denmark, whose country currently holds the Council presidency, called the vote 'a defining moment for democratic governance of technology'. Nordic ministers have already signalled plans to establish a joint AI supervisory hub in Helsinki to coordinate enforcement across the five Nordic member states.