US-China AI race drives new containment strategies

The US-China AI race isn’t about who develops advanced AI first, but rather preventing the opponent from ever reaching certain capabilities. This containment-focused approach requires verifiable agreements that one side has abandoned development efforts—an understudied area that demands urgent attention as both nations have mutual interest in preventing the other from developing certain AI capabilities that could threaten national security.

The big picture: The competition between the US and China over AI development is better understood as a containment game rather than a race, requiring verification mechanisms to ensure neither side develops certain dangerous capabilities.

• Even if the US develops safe and powerful AI systems first, the competition remains unresolved as long as China continues pursuing similar technology.
• The stakes are particularly high when considering AI systems that could potentially disrupt nuclear command and control systems.

Why containment matters: Certain AI capabilities present such significant risks that both countries have an interest in preventing the other from developing them.

• It’s insufficient for one country to simply develop AI first if the other country can eventually catch up or surpass them.
• The race dynamic incentivizes cutting corners on safety as countries rush to develop capabilities, potentially leading to loss of control over AI systems.

The verification challenge: Methods for confirming that a country isn’t secretly developing advanced AI are severely understudied.

• Creating reliable “verifiable non-development” systems and treaties is essential for any successful containment strategy.
• These verification methods could also help slow or prevent the race entirely if countries can confirm others aren’t actively developing certain AI capabilities.

Practical realities: Given perceived high stakes, China might only yield on AGI development if the US establishes a truly dominant lead.

• Such dominance is far from guaranteed, even if the US believes it could ultimately win a direct race.
• Both nations would benefit from reducing competition stakes through collaborative safety standards and making critical systems more resistant to AI-based disruption.

Reading between the lines: The author’s previous role on OpenAI’s AGI Readiness team suggests these concerns reflect thinking within organizations at the forefront of advanced AI development.

• The framing of this issue as a containment problem rather than a race problem represents an important evolution in how AI competition should be understood.
• The emphasis on verification mechanisms points to a potential diplomatic pathway that hasn’t received sufficient attention from policymakers.