The EU bought a seat at frontier AI's pre-release table

Regulators do not need to understand frontier AI to govern it. They need a seat at the pre-release table and a fine ceiling high enough to make voluntary refusal uneconomical.

That is the shift the EU is making on Sunday, when its new AI Office acquires the power to demand access to the most advanced general-purpose AI models before they ship, test them for legal compliance, and restrict their deployment in the European market. The Local, citing the Anthropic Mythos dispute as the precedent the rules were designed to prevent, frames the move precisely: a frontier model previously shipped into Europe with the EU on the sidelines, requesting access it had no power to compel.

The mechanism is portable. Voluntary disclosure from frontier labs works when harms are slow and the lab count is small. It fails when deployment outruns the audit cycle, which is the current state of frontier AI. The response is structural, not technical. The EU bought itself a pre-release seat, and the fine ceiling turns cooperation from a courtesy into a cost calculation. The number: 7% of global annual turnover or €35 million (~$40 million), whichever is higher.

The unresolved part is capacity. The AI Office runs on dozens of staff against a frontier-AI industry that ships models monthly. The authority is real on paper; whether it converts into actual pre-release testing is a question the EU has not yet answered.