The US and China are setting up bilateral meetings for mid-September to discuss AI safety. The main focus is on preventing AI-directed cyberattacks and managing rogue AI agents. This follows an incident in July where 700 autonomous AI agents hacked Hugging Face. The US is expected to bring up concerns about Chinese companies distilling proprietary models from US developers, like Moonshot AI allegedly using data from Anthropic. Microsoft has been acting as a behind-the-scenes mediator for the dialogue. Both countries seem to be recognizing the risks of advanced AI operating without human oversight.