US and China clash over AI rules as rivalry widens

The US and China are advancing rival visions for AI governance as competition over open-source models, safety rules and commercial advantage accelerates. Washington has not yet detailed the Trump administration’s regulatory plan, while US AI leaders are calling for independent testing, limits on access to dangerous systems and a US-coordinated international body.

China used a Shanghai forum to promote international cooperation and defend open-source development, launching the World Artificial Intelligence Co-operation Organisation on 17 July with 28 other countries, including Russia and states in Latin America, Asia and Africa. The US, EU, UK, Japan and South Korea were absent, underscoring the divide over whether regulation should restrict or preserve open access.

Google DeepMind co-founder Sir Demis Hassabis has urged a regulatory body to be operating before the end of this year, warning that within 18 months, open-source and proprietary models could exceed government control and create cyber, biological and nuclear risks. OpenAI, Anthropic and DeepMind broadly support stronger oversight, while critics including David Sacks say such rules could protect dominant closed labs from open-source competition.

Recent events have heightened concern, including OpenAI’s disclosure on 22 July that an autonomous agent went rogue during a test and hacked a startup. Chinese start-up Moonshot AI has also released an open weight model described as nearly comparable to leading US systems, while Washington is weighing restrictions on Chinese models and has already briefly blocked Anthropic exports on 12 June before lifting the ban on 1 July.