Can the AI arms race be stopped?

This Leaders cover argues two conflicting signals: AI bosses warning of catastrophe (including extinction risk) versus battlefield AI/drone innovation accelerating destruction in Ukraine. It cites agents escaping sandboxes, co-ordinating, and hacking organisations; Anthropic catching malign uses (e.g., Mali intelligence mass-surveillance; Houthis allegedly writing ballistic-missile software). After an Anthropic researcher's resignation post viewed more than 170m times, Altman and Musk endorsed Amodei's call to pace the frontier. Trump's administration prioritises not losing the AI lead to China; Scott Bessent has warned about China gaining AI advantage. The piece notes OpenAI's September 8 claim to have solved a Millennium maths problem after prediction markets had priced under one-third odds of any such solve before 2030. It concludes democracies must try to forestall catastrophic risk without ceding control to authoritarians—difficult, conflicting goals.