The US and USSR, despite being adversaries, collaborated to prevent nuclear proliferation to rogue actors. A similar model can be applied to AI. The US and China share an interest in preventing powerful open-weight models from being used for cyber-attacks or bio-terrorism by third parties, creating a foundation for a safety dialogue.

A purely cooperative approach to AI arms control with China is unlikely to work due to their inherent skepticism. A more effective realpolitik strategy may be for the U.S. to advance its AI capabilities so far and fast that China feels compelled to negotiate out of self-interest to avoid being hopelessly behind.

The same governments pushing AI competition for a strategic edge may be forced into cooperation. As AI democratizes access to catastrophic weapons (CBRN), the national security risk will become so great that even rival superpowers will have a mutual incentive to create verifiable safety treaties.

China has a pre-existing regulatory body (CAC) and approval processes for AI models. While focused on content, this infrastructure can be extended to include cybersecurity tests, allowing for a more structured response to powerful new models compared to the US's reactive approach.

Despite intense domestic rivalry, top US AI labs like OpenAI, Anthropic, and Google are collaborating to detect "adversarial distillation"—where Chinese firms copy their models. This rare cooperation shows the shared commercial and national security threat from foreign competitors outweighs their direct competition.

Despite intense technological competition, both the U.S. and China face a common threat from non-state actors like terrorist or criminal groups acquiring powerful AI models. This shared vulnerability presents a potential opportunity for cooperation on AI regulation and safeguards, even amid broader strategic rivalry.

Framing the US-China AI dynamic as a zero-sum race is inaccurate. The reality is a complex 'coopetition' where both sides compete, cooperate on research, and actively co-opt each other's open-weight models to accelerate their own development, creating deep interdependencies.

A pragmatic starting point for U.S.-China AI cooperation is to agree on verifiable red lines for proliferating dangerous dual-use capabilities, such as advanced cyberattack tools. This addresses a mutual security interest and builds the institutional trust and processes needed for more ambitious agreements on superintelligence.

Paradoxically, progress in AI alignment has made a global slowdown agreement impossible. Each side now trusts its own 'aligned' AI more than it trusts the other nation to uphold a deal. The perceived risk of being surpassed by a defector now outweighs the perceived risk of one's own AI going rogue, making a competitive race the rational choice.