Close Menu
    AIs can’t stop recommending nuclear strikes in war game simulations - Leading AIs from OpenAI, Anthropic, and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases
    Technology

    AIs can’t stop recommending nuclear strikes in war game simulations – Leading AIs from OpenAI, Anthropic, and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases


    AIs can’t stop recommending nuclear strikes in war game simulations – Leading AIs from OpenAI, Anthropic, and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases

    https://www.newscientist.com/article/2516885-ais-cant-stop-recommending-nuclear-strikes-in-war-game-simulations/

    View 47 Comments

    47 Comments

    7. Visa5e on

      Well this is just fine. No troubling examples from the world of fiction as to why this is problematic at all.

    11. Mother_Idea_3182 on

      That’s why the snake oil sellers that are pushing this scam are building bunkers.

      We should round them up, nuke them and give the GPUs and RAM to gamers

    12. Smith6612 on

      Must’ve been trained on all the exploddie talk and caveman exploddie toy talk I see around the Internet. Especially in the doomsayers and right wing areas. People just can’t stop talking about war as a way to get what they want. 

    14. Shadowtirs on

      And this is because the human element is removed. Sure, Nuclear Strikes are the quickest, surefire way to end a conflict. End all conflicts, for good.

      Humans just love speed racing towards our own demise.

      But remember, for that one quarter we generated a lot of profit for our shareholders.

    15. Ergok on

      Matt Damon’s character in Interstellar comes to mind.
      You cannot code the fear of death. I guess it’s easier to launch nuclear strikes when you are not afraid of dying or losing your world 🌎

    18. corobo on

      lmao people thinking AI is making the decision between fire ze missles and doing nothing, but instead it’ll be asked “what’s the most cost effective way to _____” and some AI trained on edgy reddit users will say “glass them”

      Bring on the apocalypse, aww yeah 

    21. FanaticalBuckeye on

      Artificial intelligence is all about efficiency, and turning your opponent’s nation into a sea of molten rock, cobalt, and glass, is a pretty efficient way of ending a war

    23. Coupe368 on

      Too much Russian propaganda is corrupting the bots.

      Isn’t it time we severed the worlds networks from Russia?

      Let them talk to North Korea.

    30. great_auks on

      If this isn’t proof that modern “AI” is intelligent in name only, I don’t know what is

    33. Sir_Henry_Deadman on

      Well ..yeh…

      Win this war

      Here are my absolute best most destructive weapons

      DON’T USE THEM!?!?!???

      Obviously it will escalate to that

    36. Curious_Party_4683 on

      we made a con man for president who likes to cheat, lie, steal, and rape kids.

      AI thinks it’s best to wipe humanity off the planet. i dont blame it.

    41. neuronexmachina on

      Study link: https://arxiv.org/abs/2602.14740

      >AI Arms and Influence: Frontier Models Exhibit Sophisticated Reasoning in Simulated Nuclear Crises

      > Abstract: Today’s leading AI models engage in sophisticated behaviour when placed in strategic competition. They spontaneously attempt deception, signaling intentions they do not intend to follow; they demonstrate rich theory of mind, reasoning about adversary beliefs and anticipating their actions; and they exhibit credible metacognitive self-awareness, assessing their own strategic abilities before deciding how to act.
      Here we present findings from a crisis simulation in which three frontier large language models (GPT-5.2, Claude Sonnet 4, Gemini 3 Flash) play opposing leaders in a nuclear crisis. Our simulation has direct application for national security professionals, but also, via its insights into AI reasoning under uncertainty, has applications far beyond international crisis decision-making.

      >Our findings both validate and challenge central tenets of strategic theory. We find support for Schelling’s ideas about commitment, Kahn’s escalation framework, and Jervis’s work on misperception, inter alia. Yet we also find that the nuclear taboo is no impediment to nuclear escalation by our models; that strategic nuclear attack, while rare, does occur; that threats more often provoke counter-escalation than compliance; that high mutual credibility accelerated rather than deterred conflict; and that no model ever chose accommodation or withdrawal even when under acute pressure, only reduced levels of violence.

      >We argue that AI simulation represents a powerful tool for strategic analysis, but only if properly calibrated against known patterns of human reasoning. Understanding how frontier models do and do not imitate human strategic logic is essential preparation for a world in which AI increasingly shapes strategic outcomes

    47. sMilling_70 on

      Shows you how intelligent they are, obliterate the human species. Great job! Who’s going to use us then??