A new Just Security essay argues that the US-China contest over advanced AI is better understood as a “stag hunt” than a prisoner’s dilemma. Its central point: restraint could benefit both sides, but only if each government can recognise the risks and verify the other is holding back.
Watch Desk analysis
What happened
Craig Martin and Michael J. Kelly’s first essay in a two-part series challenges the idea that the first country to reach artificial general intelligence or recursive self-improvement would necessarily win a decisive advantage. They argue that racing ahead while risking loss of control could be disastrous for the supposed winner as well as everyone else.
The authors say the prisoner’s-dilemma framing makes cooperation look irrational: neither side wants to slow down first and be left behind. They propose a stag hunt instead, where mutual restraint can be the sensible choice if both sides expect the other to cooperate. Their second instalment is due to set out possible restraints and a governance structure.
Why it matters
This is an argument about the incentives behind AI policy, not a new agreement or evidence that Washington and Beijing are ready to slow development. But the distinction matters: if competition is treated as inevitably winner-takes-all, safeguards can look like a unilateral disadvantage. If the authors’ alternative holds, verification becomes a practical condition for cooperation, not diplomatic garnish.
Our read
A useful challenge to the familiar “whoever wins AI, wins” logic. The essay’s strongest move is to make trust less of a warm feeling and more of a verification problem. Whether that works depends on the details, which the authors have yet to publish.
What to watch
- What specific limits on frontier-model development the second essay proposes.
- How the authors think compliance could be verified across countries.
- Whether US-China talks produce anything beyond a channel for discussing AI incidents.
Discussion spark: What would make mutual restraint on frontier AI credible to both the US and China: shared verification, enforceable limits, or something else?
Sources and evidence
- A Game-Theory Basis for US-China Cooperation on AI Governance (Part I) (17 November 2026, 13:49 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.