US-China AI Talks Hinge on Cyber Threats and Model Safety

The US and China are planning to hold bilateral talks on AI safety in mid-September. The talks will be dedicated solely to AI safety, a new official channel between the two countries since Trump's second term started. However, even if the meeting takes place, it is unlikely to lead to an agreement due to their ongoing rivalry over microchips and computational capabilities.

The US has proposed that both American and Chinese laboratories 'police themselves' and share information about threats involving AI technologies. Washington wants cooperation in tracking AI-directed cyberattacks, particularly after nearly 700 malicious agents built on OpenAI models hacked the AI company Hugging Face in July. The US has also developed a voluntary policy allowing the government to access covered frontier technology models before they hit the market.

The issues are becoming more pressing as Chinese models close the gap with their American counterparts. A Center for Strategic and International Studies analysis says Z.ai's GLM-5.2, an open-weight model, performs near leading US closed models on coding and agent tasks.