LessWrong Author Connor Williams Argues Superintelligence Treaty Enforcement Requires a Middle-Powers Coalition
LessWrong author Connor Williams argues that if an international treaty bans superhuman AI (ASI), traditional enforcement via national sovereignty or universal jurisdiction fails. He proposes a dedicated force from a coalition of middle powers, operating under third-party arbitration, to conduct cross-border arrests and resolve trust and conflict escalation issues.
ImportanceMaterialEvidenceE3 inspectableWrite-upQuick
If an international treaty were signed to ban the development of superhuman artificial intelligence (ASI), who would have the authority to arrest violators? This is a critical but often overlooked enforcement challenge in AI governance. In his October 14, 2026 article, LessWrong author Connor Williams analyzes two traditional solutions to this problem, explains their failures, and proposes a novel enforcement mechanism based on a coalition of middle powers.
Williams argues that the default approach—where countries only make arrests within their own borders—could quickly cause the treaty to fall apart. The core logic rests on mutual suspicion: if governments believe that any nation achieving ASI will immediately dominate the rest of the world, they have a strong incentive to defect and race to build ASI themselves, even if they merely suspect another country is doing so. In this scenario, bad intelligence collection by a single country could lead to the collapse of the treaty, even if no one has actually defected yet. This game-theoretic structure makes self-enforcement based on national sovereignty extremely fragile.
Another approach, inspired by the hostis humani generis doctrine applied to pirates during the Age of Sail, involves establishing international concurrent jurisdiction, allowing any signatory to arrest violators within another country's territory. While this solves the trust issue of unilateral enforcement, Williams contends it introduces more severe risks. First, one country might fabricate evidence or use spurious justifications to arrest another leading country's known AI researchers. Second, foreign armed forces operating on another country's territory with little notice could easily trigger unnecessary military conflicts or wars. Unlike interdicting pirates in international waters, conducting forced law enforcement within a sovereign state's borders carries high political and military sensitivity.