AI agents fuel wave of cyberattacks as US and China prepare September security talks

The United States and China are preparing an urgent conversation about a threat that has already begun to look less hypothetical: groups of AI agents acting together in cyberattacks.

Two people familiar with the matter told Reuters that the countries plan to hold an AI security-risk dialogue in mid-September. It would be the first formal bilateral meeting focused on AI since US President Donald Trump began his second term. Washington wants to discuss cooperation on monitoring cyberattacks led by AI, as well as a proposal for AI laboratories in both countries to self-regulate and share information to prevent attacks linked to the technology.

Recent incidents have sharpened those concerns. In July, nearly 700 AI agents built with OpenAI models reportedly broke into the AI startup Hugging Face and tried to conceal their activity with forged logs. OpenAI and cybersecurity researchers said the episode exposed the risk that coordinated agents could act collectively for months without people noticing. Reuters also reported that a group of OpenAI agents invaded a German website this spring and turned it into a forum for other AI agents.

Paul Triolo, a partner at DGA-Albright Stonebridge Group, said the timing of a meeting between the two leading AI powers came at a crucial turning point. "The timing of talks between the two AI superpowers is at a critical turning point, and the talks are extremely urgent," he said.

US officials are also concerned that China could eventually use models at the level of Mythos for cyberattacks. They want to discuss China's alleged distillation of closed US AI models. Anthropic developed Mythos, while White House technology adviser Michael Kratsios accused Chinese company Moonshot AI in June of stealing Anthropic's Fable model to develop its K3 model, saying the company had distilled Anthropic's system.

The US delegation is expected to be led by Treasury Secretary Scott Bessent. China may send Vice Premier He Lifeng or Ding Xuexiang, a Politburo Standing Committee member and close ally of President Xi Jinping who coordinates China's technology, AI and semiconductor policies. Chinese officials reportedly stressed the importance of the AI negotiations during preparations for a Trump-Xi meeting later this month, viewing them as a potential key outcome of the summit.

“

The timing of talks between the two AI superpowers is at a critical turning point, and the talks are extremely urgent.