America can hit an AI “kill switch.” Europe must switch gears before it’s too late

America’s ability to restrict access to leading AI systems—whether through export controls, licensing, or cybersecurity-driven directives—turns frontier AI into a geopolitical lever. Even if the stated reasons are framed as risk management, the practical effect is the same: European organisations that depend on cutting-edge models can face abrupt, operational disruptions, just when those systems are becoming embedded in everything from research workflows to cybersecurity operations.

That is why the “kill switch” framing lands. It captures a reality Europe has been trying to avoid: the most advanced AI capability is not simply a technological advantage, but also a dependency. When critical tools are supplied from outside Europe, control over availability is outside Europe too. And once advanced AI becomes a routine input to national resilience—cyber defence, innovation pipelines, industrial automation, and scientific discovery—interruption stops being an inconvenience and starts being a strategic risk.

Europe’s response shouldn’t be panic or symbolic posturing. The real “switch gears” moment is about capability and leverage. Europe must treat AI sovereignty as an engineering and investment problem, not just a slogan: scale advanced computing, expand talent and research, and strengthen the digital infrastructure that allows local development and safe deployment. That means building more than research papers and pilots; it requires the capacity to train, evaluate, and operate models under European constraints and governance expectations.

Europe also needs redundancy and resilience. Even if some US-controlled systems remain useful and lawful to use, dependency must be reduced through alternatives—whether via open-weight models, European providers, or partnerships that keep key risks manageable. For scientific and cybersecurity communities in particular, a “plan B” should be treated as operational hygiene: if the best tool becomes unavailable, the workflow cannot collapse.

Finally, Europe has to make governance enforceable. The goal is not only to regulate AI on paper, but to ensure systems used in Europe meet safety, auditability, and security standards. When rules are paired with technical requirements and incident response pathways, the risk of sudden shocks—whether from supplier decisions or malicious misuse—shrinks.