EU seeks global safety rules despite US-China divide

The EU is pushing international agreements on frontier AI safety despite limited prospects of securing binding commitments from the United States and China. Europe trails both powers in developing advanced models but sees its regulatory experience as a source of influence. Recent diplomatic initiatives have backed human oversight, mandatory testing before deployment and international supervision, although most commitments remain symbolic and lack enforcement mechanisms.

Proposed safeguards include independent model evaluations, stronger oversight for the most powerful systems and an international body that could authorize pauses in development. The Council of Europe’s Framework Convention on AI and Human Rights offers a treaty framework, but the United States has not ratified it. Washington has resisted international commitments, while China has emphasized access and economic development over safety. Industry positions also differ: Google DeepMind and Anthropic leaders have supported international standards or pacing development, while OpenAI has warned against slowing American companies relative to competitors.

The Geneva AI Safety Convention in June 2027 offers another opportunity to negotiate minimum standards for risky models. A potential alternative is voluntary cooperation between the EU and countries such as Canada and the United Kingdom. Shared testing requirements could influence companies seeking customers beyond the United States and China, giving smaller economies greater collective leverage.