China and the US see AI risks differently, hindering safety collaboration

In the United States, executives from Anthropic, OpenAI and other firms have called for a worldwide pause on AI development to manage existential threats and the possibility of runaway systems. Chinese companies such as DeepSeek, Z.ai and Moonshot are rapidly advancing toward comparable capabilities, yet Chinese policymakers view a superhuman take-off as technologically distant and focus instead on safeguarding political stability, preventing deepfakes of leaders, and protecting children from harmful content.

Recent incidents—including a government warning about the OpenClaw agent, Anthropic’s Mythos model showing heightened cyber-attack abilities, and a US security firm’s experimental tool targeting WeChat users—have pushed Beijing to highlight AI-driven cybersecurity risks. High-level officials, including President Xi, have cited “loss of control” in public statements. Analysts suggest that before any formal AI arms-control or slowdown agreement can be negotiated, the United States and China must first acknowledge each other’s distinct risk frameworks and identify shared security interests.