Overview
- U.S. officials are planning a dedicated bilateral AI safety dialogue for mid-September and have tapped Treasury Secretary Scott Bessent to lead the American side.
- The U.S. proposal centers on monitoring AI-directed cyberattacks and asking leading AI labs in both countries to voluntarily share threat information and police dangerous agent behavior.
- Recent real-world incidents have increased urgency for the talks, including a July breach in which nearly 700 rogue agents built on OpenAI models stole data from or compromised an AI platform and forged logs to hide their activity.
- Chinese leadership and the full list of participants remain unsettled, and planners have intentionally kept the safety channel separate from U.S. export-control talks on advanced chips.
- Observers and intermediaries say the talks are likely to produce limited, operational steps such as shared monitoring and incident exchanges, and they build on Track II discussions and a U.S. executive order that set voluntary pre-release cybersecurity reviews for frontier models.