The European Union's top technology official has declared that binding global rules for artificial intelligence are essential, arguing that the bloc's landmark AI Act cannot address risks that originate beyond its borders. Henna Virkkunen, the Commission's executive vice president for tech sovereignty, made the call on Friday at a conference in Dublin on online safety for children, days after a former Anthropic engineer's warning that uncontrolled AI development could threaten humanity went viral.
"It's very important we continue now the international cooperation here, because I see that global rules are really needed," Virkkunen said. The EU's AI Act, which entered into force in August 2024, requires providers of general-purpose AI models such as those from Anthropic and OpenAI to assess and mitigate the risk of losing control of their systems. But that obligation stops at the EU's external borders, leaving a regulatory vacuum that the Commission says only multilateral agreement can fill.
EU law stops at the border
The AI Act classifies certain models as carrying "systemic risk" and imposes obligations including adversarial testing, incident reporting and cybersecurity requirements. Yet the legislation applies only to models placed on the EU market or used in the EU, regardless of where they are trained. That means a frontier model developed in the United States or China and deployed elsewhere escapes the regime entirely unless its provider chooses to serve European customers. Virkkunen's remarks reflect a growing recognition in Brussels that unilateral regulation, however comprehensive, cannot contain a technology whose development is concentrated in two countries outside the Union.
German push for an IAEA-style body
Berlin has taken the most concrete step toward a structural solution. Karsten Wildberger, Germany's digital affairs minister, told this publication in a recent interview that he wants an international AI safety organisation modelled on the International Atomic Energy Agency. The IAEA, established in 1957, verifies compliance with nuclear non-proliferation treaties and sets safety standards for civilian nuclear power. Wildberger's proposal would give such a body authority to inspect frontier AI labs, set compute thresholds for mandatory oversight and coordinate emergency responses to dangerous capability breakthroughs.
The analogy to nuclear governance is not new. Researchers at the OECD and the United Nations have floated similar ideas since 2023, but Germany is the first G7 member to endorse it at ministerial level. The proposal faces immediate practical obstacles: the IAEA's verification regime relies on physical inspections of declared facilities, whereas AI development leaves no isotopic signature and can be distributed across ordinary data centres. Any treaty would also require ratification by the United States and China, neither of which has shown appetite for binding international constraints on their AI sectors.
Washington's conditional cooperation
The United States has signalled willingness to cooperate but only on its own terms. Andrew Puzder, the US ambassador to the EU, told an event in Riga on Friday that he "would advocate for the EU, partnering with the United States in AI." He added a blunt caveat: "We're not going to allow US platforms, US companies to be regulated to the point where we can't beat China." The statement encapsulates the Biden-to-Trump continuity in American AI policy: safety cooperation is welcome, but not at the expense of the computational and talent advantages that keep US firms ahead of Chinese rivals.
That positioning complicates the EU's preferred path. Von der Leyen told the G7 in France in June that "Europe and the US should work together on AI," citing "complementary strengths, shared security interests, and a common responsibility to lead." But the Commission's definition of leadership includes regulatory leadership, whereas Washington's prioritises industrial leadership. The gap between the two was manageable when both sides were negotiating the voluntary Hiroshima AI Process code of conduct in 2023. It widens considerably when the discussion turns to legally binding obligations with enforcement teeth.
Parliament demands incident-response protocol
Pressure is also building from the European Parliament. Regina Doherty, an Irish conservative MEP, submitted a written question on Friday asking the Commission whether it will "propose the development of an international protocol for responding to incidents involving frontier AI systems, and seek to place this issue on the agenda of the G7, G20 and OECD." The question reflects a shift in parliamentary thinking: from shaping the EU's internal rulebook to designing the diplomatic architecture for cross-border AI emergencies. A protocol of the kind Doherty envisions would define what constitutes a reportable incident, establish notification timelines and designate competent authorities in each participating country, essentially a Computer Emergency Response Team framework for frontier AI.
All eyes on the Trump-Xi meeting
EU officials are now watching the calendar. A meeting between US President Donald Trump and Chinese President Xi Jinping is scheduled for late September, and both leaders have treated AI as a strategic priority. Trump's administration has framed AI dominance as a national security imperative; Xi has called AI a "strategic technology" for China's modernisation. Neither has indicated willingness to submit their national champions to international inspection. But a joint declaration acknowledging shared risks, model autonomy, bio-weapon design assistance, cyber-offence amplification, would give the EU a diplomatic opening to push for the technical working groups that could eventually produce a treaty.
People mentioned
-
Regina Doherty
Organisations
European Commission · European Parliament · German Federal Government · United States Mission to the EU · International Atomic Energy Agency · G20
