For global AI governance, the US and China must cooperate while competing

The call for US-China collaboration on AI governance arrives at a pivotal moment, when the technology’s capabilities are outpacing the frameworks meant to contain its risks. Anthropic’s cautious deployment of Fable and Mythos—prompted by cybersecurity threats—exposes a fundamental truth: no single corporation, however responsible, can shoulder the burden of safeguarding a technology with global implications. The private sector lacks the authority, the enforcement tools, and the cross-border mandate to address threats that do not respect national boundaries. This is not an argument against innovation, but a recognition that governance must be elevated to the intergovernmental level. China’s rapid progress with open-weight models complicates the picture. On one hand, it demonstrates that the United States no longer holds a monopoly on cutting-edge AI, which could erode Washington’s strategic advantage. On the other, it creates a rare opportunity for mutual learning: Chinese researchers have made significant strides in efficiency and accessibility, while American firms lead in frontier safety research. A partnership built on shared technical standards—for testing, transparency, and incident reporting—could benefit both nations and the world at large. Yet such collaboration is fraught with geopolitical tension, as each side views the other’s AI ambitions through a lens of national security and economic competition. The urgency is real. Uncoordinated development risks a race to the bottom, where speed is prioritized over safety, and where malicious actors exploit gaps between jurisdictions. A bilateral framework, even if imperfect, could establish norms for responsible AI deployment, create early-warning systems for catastrophic failures, and prevent the weaponization of autonomous systems. Skeptics will argue that China cannot be trusted to adhere to shared rules, and that any technology transfer would only strengthen an adversary. But the alternative—complete decoupling—would fragment the global AI ecosystem, making oversight even harder and driving innovation underground. Ultimately, the question is not whether the US and China should cooperate, but how to do so without sacrificing their respective interests. The answer may lie in compartmentalized agreements: focusing on non-military applications, humanitarian uses, and safety research, while leaving defense-related AI out of scope. The Anthropic incident is a warning shot—a reminder that the risks are too large for any single nation to manage alone. If Washington and Beijing can find common ground on the fundamentals of AI safety, they might not just mitigate dangers, but also build a foundation for broader diplomatic trust in an increasingly digital century.