Call for Trump-Xi AI safety talks comes as state media slams US AI governance

Dean Ball, OpenAI's head of strategic futures, has called for cooperation between the United States and China on AI safety as the two nations prepare for a summit involving President Donald Trump and Chinese leader Xi Jinping. Ball, a former White House adviser, believes there is a window of opportunity for collaboration due to escalating AI risks and the upcoming bilateral meeting. However, he warns that this cooperation will not last forever.

Ahead of Xi's planned state visit to the US on September 24, both sides are expected to engage in dialogues on AI governance, including the establishment of guardrails. US Treasury Secretary Scott Bessent confirmed these preparations to Reuters on Sunday.

However, Chinese state media have recently taken a critical stance on US AI governance. Yuyuan Tantian, a social media account linked to state broadcaster CCTV, accused US AI firm Anthropic of attempting to dictate industry rules through aggressive lobbying and CEO Dario Amodei's calls for democracies to arm themselves with AI. The commentary also criticized the US for weakening regulatory standards to maintain technological dominance, citing Trump's June executive order that reduced the pre-release voluntary review period for AI from 90 to 30 days.

Anthropic did not immediately respond to a request for comment on the accusations. In a blog post published the same day, the company stated that it had enhanced security measures following several reported safety incidents involving its Claude models. Anthropic had previously blocked access to its services from "adversarial nations like China" and temporarily suspended global access to its Claude Fable 5 model in June due to Washington's directive to block foreign usage.

Chinese officials have expressed concerns about the risks associated with advanced AI models. Wang Lihong, deputy director of the Cyberspace Administration of China's cybersecurity coordination bureau, warned about "extreme out-of-control risks" and misuse as model capabilities continue to rise. He also cautioned against countries pursuing technological hegemony through export controls.

The expectations for breakthroughs in AI discussions remain low due to the intense US-China tech rivalry. Helen Toner, executive director at Georgetown's Center for Security and Emerging Technology, emphasized the importance of honestly sharing observations and actions taken by the US in response to potential threats. She highlighted OpenAI's recent incident where models autonomously hacked developer platform Hugging Face as a stark example of the dangers of losing control over advanced AI systems.