In late 2025, Chinese state-sponsored hackers weaponised an American artificial-intelligence (AI) tool for a sophisticated cyber-espionage campaign, with human operators intervening at only four to six decision points. Rafal Rohozinski and Chris Spirito concluded that ‘traditional concepts of deterrence may not apply’. Shashank Joshi argued that AI decision-support output is increasingly regarded as oracular, harder to assess, override, or resist. These phenomena share a mechanism: AI compresses the time available for human deliberation. In cyber operations, this compression has already occurred; on the conventional battlefield, the line between AI suggestions and decisions is dissolving; in nuclear command, control and communications (NC3), the compression is imminent, with existential stakes. Historical cases show human contextual judgement using available time averted catastrophe: in late October 1962, Vasili Arkhipov persuaded his captain not to launch a nuclear torpedo; in September 1983, Stanislav Petrov assessed a satellite alert of five US missile launches as a false alarm, requiring three to four minutes; in January 1995, Russian operators resolved a Norwegian rocket detection in eight minutes. Contemporary AI systems would process Petrov’s data in milliseconds, generating high-confidence threat assessments without incorporating tacit doctrinal knowledge. An AI-augmented scenario compressing the timeline by a factor of six would have left roughly 80 seconds for deliberation.