AI superintelligence could shatter nuclear deterrence, warns LessWrong essay
How recursive AI improvement might make second-strike guarantees obsolete.
In a provocative crosspost on LessWrong, Felix Choussat examines how superintelligent AI could fundamentally undermine nuclear deterrence. He notes that past military technologies—ICBMs, MIRVs, thermonuclear staging—were introduced slowly enough for rivals to build second-strike capabilities. However, AI's recursive self-improvement will accelerate R&D by years or centuries, creating a decisive strategic gap.
Choussat identifies three pathways for an AI-empowered state to negate nuclear deterrents: (1) a 'splendid first strike' that locates and destroys all enemy counterforce (including SSBNs) via autonomous sensors; (2) advanced WMD defenses neutralizing retaliatory strikes; (3) escalation management through salami-slicing or manipulation to prevent launch. He concludes that without parallel AI investment, even nuclear powers will eventually lose their deterrent edge.
- Historical military innovations were slow enough for states to adapt; AI promises a technological sprint that outpaces second-strike guarantees.
- Three AI-enabled pathways to overcome nuclear deterrence: locating all counterforce, deploying missile defenses, or manipulating enemy decision-making.
- Without comparable AI bases, nuclear states risk complete strategic disempowerment despite their arsenals.
Why It Matters
AI could render nuclear arsenals irrelevant, rewriting global power dynamics and strategic stability.