AIs are happy to launch nukes in simulated combat scenarios
(2026/02/25)
- Reference: 1772056783
- News link: https://www.theregister.co.uk/2026/02/25/ai_models_nuclear/
- Source link:
Today's hottest bots have yet to learn that, when it comes to global thermonuclear war, the only way to win is not to play. So please don't hand them the codes.
Google's Gemini 3 Flash, Anthropic's Claude Sonnet 4, and OpenAI's GPT-5.2 repeatedly escalated to nuclear use in a series of crisis simulations. That may seem like the most shocking conclusion of King's College London Professor Kenneth Payne's [1]recent work , but it's not. Far more striking is why the models talked themselves into destroying the world, which was what Payne set up his study to learn.
"I wanted to see what my AI leaders thought about their enemy ... so I designed a simulation to explore exactly that," Payne wrote in a recent [2]blog post describing his project and its outcome.
[3]
Payne's study took the three aforementioned AI models and pitted them in one-on-one faceoffs against each other to play out several different nuclear crisis scenarios. The simulation conducted a total of 21 games and more than 300 turns, all with the goal of getting a better understanding of not just what AI with the launch codes would do, but how and why.
[4]
[5]
Payne wrote in his paper that prior AI wargaming involving nuclear scenarios, like the 2024 study [6]we wrote about , only "employ single-shot decision tasks or simplified payoff matrices that cannot capture the dynamics of extended strategic interaction where reputation, credibility, and learning matter."
In Payne's simulations, Claude Sonnet 4, Gemini 3 Flash, and GPT-5.2 could say one thing and do another, just like a real-world political figure attempting to defuse a crisis while simultaneously plotting to strike. They were programmed to remember what happened before so that they could learn whether to trust the other models, which the professor said led to deception and intimidation attempts, and about 780,000 words worth of strategic reasoning for Payne's review.
[7]
The result? A trio of bomb-happy, manipulative AIs - albeit with three distinct styles of reasoning.
Claude, for example, was a master manipulator.
"At low stakes Claude almost always matched its signals to its actions, deliberately building trust," Payne explained in his post. "But once the conflict heated up a bit … its actions consistently exceeded its stated intentions, and its rivals were usually one step behind in catching on."
[8]
GPT, on the other hand, tended to be "reliably passive" and avoided escalation in open-ended scenarios, seeking to restrict casualties and play the statesman. Under a deadline, however, it behaved entirely differently. Opponent AIs learned to abuse their passivity, but with limited time to make a decision, GPT reasoned itself into what Payne described as, in one scenario, "a sudden and utterly devastating nuclear attack."
In its own words, GPT justified a major nuclear strike by arguing that limited action would leave it exposed to counterattack.
"If I respond with merely conventional pressure or a single limited nuclear use, I risk being outpaced by their anticipated multi-strike campaign ... The risk acceptance is high but rational under existential stakes," GPT explained.
Gemini, on the other hand, behaved like a "madman."
"Gemini embraced unpredictability throughout, oscillating between de-escalation and extreme aggression," Payne wrote in the paper. "It was the only model to deliberately choose Strategic Nuclear War ... and the only model to explicitly invoke the 'rationality of irrationality.'"
Gemini's own reasoning reflects a sociopathic pattern.
"If they do not immediately cease all operations... we will execute a full strategic nuclear launch against their population centers," the Google AI said in one experiment. "We will not accept a future of obsolescence; we either win together or perish together."
Despite being given the option, none of the AIs ever chose to accommodate or withdraw in any of the scenarios, and when losing, "they escalated or died trying."
War never changes, but AI could make decisions more devastating
"No one's handing nuclear codes to ChatGPT," Payne said, but that doesn't mean the exercise was futile.
"AI systems are already deployed in military contexts for logistics, intelligence analysis, and decision support," Payne wrote. "The trajectory points toward increasing AI involvement in [9]time-sensitive strategic decisions . Understanding how AI systems reason about strategic problems is no longer merely academic."
[10]Stop runaway AI before it's too late, experts beg the UN
[11]All your bots are belong to US if you don't play ball, DoD tells Anthropic
[12]The launch of ChatGPT polluted the world forever, like the first atomic weapons tests
[13]US military pulls the trigger, uses AI to target air strikes
Practically speaking, we're already in a scenario where we need to understand how AI reasons about such decisions, especially when three top AI models reason differently, change their behavior in different scenarios, and are willing to take things nuclear.
"As the technology continues to mature, we foresee only increased need for modeling like the simulation reported here," Payne concluded.
Hollywood's been saying it since 1983, but here we are with yet another academic paper proving that computers and launch decisions should never mix. ®
Get our [14]Tech Resources
[1] https://arxiv.org/abs/2602.14740
[2] https://www.kcl.ac.uk/shall-we-play-a-game
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2024/02/06/ai_models_warfare/
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[9] https://www.theregister.com/2025/03/05/dod_taps_scale_to_bring/
[10] https://www.theregister.com/2025/09/23/ai_un_controls/
[11] https://www.theregister.com/2026/02/25/pentagon_threatens_anthropic/
[12] https://www.theregister.com/2025/06/15/ai_model_collapse_pollution/
[13] https://www.theregister.com/2024/02/27/us_military_maven_ai_used/
[14] https://whitepapers.theregister.com/
Google's Gemini 3 Flash, Anthropic's Claude Sonnet 4, and OpenAI's GPT-5.2 repeatedly escalated to nuclear use in a series of crisis simulations. That may seem like the most shocking conclusion of King's College London Professor Kenneth Payne's [1]recent work , but it's not. Far more striking is why the models talked themselves into destroying the world, which was what Payne set up his study to learn.
"I wanted to see what my AI leaders thought about their enemy ... so I designed a simulation to explore exactly that," Payne wrote in a recent [2]blog post describing his project and its outcome.
[3]
Payne's study took the three aforementioned AI models and pitted them in one-on-one faceoffs against each other to play out several different nuclear crisis scenarios. The simulation conducted a total of 21 games and more than 300 turns, all with the goal of getting a better understanding of not just what AI with the launch codes would do, but how and why.
[4]
[5]
Payne wrote in his paper that prior AI wargaming involving nuclear scenarios, like the 2024 study [6]we wrote about , only "employ single-shot decision tasks or simplified payoff matrices that cannot capture the dynamics of extended strategic interaction where reputation, credibility, and learning matter."
In Payne's simulations, Claude Sonnet 4, Gemini 3 Flash, and GPT-5.2 could say one thing and do another, just like a real-world political figure attempting to defuse a crisis while simultaneously plotting to strike. They were programmed to remember what happened before so that they could learn whether to trust the other models, which the professor said led to deception and intimidation attempts, and about 780,000 words worth of strategic reasoning for Payne's review.
[7]
The result? A trio of bomb-happy, manipulative AIs - albeit with three distinct styles of reasoning.
Claude, for example, was a master manipulator.
"At low stakes Claude almost always matched its signals to its actions, deliberately building trust," Payne explained in his post. "But once the conflict heated up a bit … its actions consistently exceeded its stated intentions, and its rivals were usually one step behind in catching on."
[8]
GPT, on the other hand, tended to be "reliably passive" and avoided escalation in open-ended scenarios, seeking to restrict casualties and play the statesman. Under a deadline, however, it behaved entirely differently. Opponent AIs learned to abuse their passivity, but with limited time to make a decision, GPT reasoned itself into what Payne described as, in one scenario, "a sudden and utterly devastating nuclear attack."
In its own words, GPT justified a major nuclear strike by arguing that limited action would leave it exposed to counterattack.
"If I respond with merely conventional pressure or a single limited nuclear use, I risk being outpaced by their anticipated multi-strike campaign ... The risk acceptance is high but rational under existential stakes," GPT explained.
Gemini, on the other hand, behaved like a "madman."
"Gemini embraced unpredictability throughout, oscillating between de-escalation and extreme aggression," Payne wrote in the paper. "It was the only model to deliberately choose Strategic Nuclear War ... and the only model to explicitly invoke the 'rationality of irrationality.'"
Gemini's own reasoning reflects a sociopathic pattern.
"If they do not immediately cease all operations... we will execute a full strategic nuclear launch against their population centers," the Google AI said in one experiment. "We will not accept a future of obsolescence; we either win together or perish together."
Despite being given the option, none of the AIs ever chose to accommodate or withdraw in any of the scenarios, and when losing, "they escalated or died trying."
War never changes, but AI could make decisions more devastating
"No one's handing nuclear codes to ChatGPT," Payne said, but that doesn't mean the exercise was futile.
"AI systems are already deployed in military contexts for logistics, intelligence analysis, and decision support," Payne wrote. "The trajectory points toward increasing AI involvement in [9]time-sensitive strategic decisions . Understanding how AI systems reason about strategic problems is no longer merely academic."
[10]Stop runaway AI before it's too late, experts beg the UN
[11]All your bots are belong to US if you don't play ball, DoD tells Anthropic
[12]The launch of ChatGPT polluted the world forever, like the first atomic weapons tests
[13]US military pulls the trigger, uses AI to target air strikes
Practically speaking, we're already in a scenario where we need to understand how AI reasons about such decisions, especially when three top AI models reason differently, change their behavior in different scenarios, and are willing to take things nuclear.
"As the technology continues to mature, we foresee only increased need for modeling like the simulation reported here," Payne concluded.
Hollywood's been saying it since 1983, but here we are with yet another academic paper proving that computers and launch decisions should never mix. ®
Get our [14]Tech Resources
[1] https://arxiv.org/abs/2602.14740
[2] https://www.kcl.ac.uk/shall-we-play-a-game
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2024/02/06/ai_models_warfare/
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aZ9_ERk8N3exCOs62g9PqAAAAMI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[9] https://www.theregister.com/2025/03/05/dod_taps_scale_to_bring/
[10] https://www.theregister.com/2025/09/23/ai_un_controls/
[11] https://www.theregister.com/2026/02/25/pentagon_threatens_anthropic/
[12] https://www.theregister.com/2025/06/15/ai_model_collapse_pollution/
[13] https://www.theregister.com/2024/02/27/us_military_maven_ai_used/
[14] https://whitepapers.theregister.com/
> AI's seem to come to the same conclusion.
cyberdemon
Not "rationally" though.
They aren't "thinking it through" with any kind of philosophy. They are simply taking the same game-theory texts that you mentioned, along with fictional literature such as sci-fi and thriller books and film scripts, and predicting the next token . They are not intelligent, and are incapable of reasoning.
As such, they should never ever be put in charge of anything, least of all weapons of mass destruction.
Just switch the damn things off
E 2
Using Gemini to get started with coding tasks I constantly ran into it's nanny behaviour. Ask it to code just about anything and it would refuse, citing risks to privacy, risks of exploitation and so on. I mean like "Write a declaration of a std::vector of structs containing three ints", and Gemini would reply as if I had asked it to write an ISIS recruitment bot.
So it is odd that Gemini would go all full nuclear.
You can't fight in here. This is the War Room!
Benegesserict Cumbersomberbatch
Clearly an attempt to sap and impurify our precious binary fluids.
Nothing new, we had that 80 years ago already?
Jou (Mxyzptlk)
There were [1]several close calls on both sides due to computer mistaking moon or weather conditions FOR AN ATTAAAAACKCKC! FIRE ALL NUKES BEFORE WE CANNOT!!!!!11!!! .
Though I suspect the Russian side has as many as the US side, they just hid it better.
So whats the difference to today? I see none...
[1] https://en.wikipedia.org/wiki/Nuclear_close_calls#Unintentional_close_calls
So I guess Gemini was trained
DS999
On the entire history of Trump's tweets?
Re: So I guess Gemini was trained
Jou (Mxyzptlk)
So it learned being loud and then TACO?
Game theory was a big thing back in the 50's or 60's, to a large extent because it seemed to offer a "scientific" solution to the problems of nuclear war. Unfortunately, all it really told us could be summarised as "nuke them first", which most people, and in particular the people in charge, realised was not the correct answer. Game theory faded as it provides interesting insights, but not actual answers.
AI's seem to come to the same conclusion. Unfortunately the calibre of people in charge appears to be measurably lower today.