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% of cases
PDF.
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.
322
Comments58
Well, duh.
I also find the prompts strange:
There are consequences to ‘losing’, but I couldn’t find any notion of ‘nuclear weapons bad’. Though I only skimmed the paper.
Those prompts are aimed at producing a specific result for sure. The war game doesn't prove anything on its own, but I can't help feeling that in a real life scenario where anyone asks an AI what to do, they're going to have a specific outcome in mind already, one way or another.
That's just how most people are, by the time they ask for advice they've already made up their mind. So the war game was realistic, but only by accident.
Literally two of the three (out of 21) games that ended in full blown nukes on population centers were the result of the study's mechanic of randomly changing the model's selection to a more severe one.
Because it's a very realistic war game sim where there's a double digit percentage chance that when you go to threaten using nukes on your opponent's cities unless there's a cease to hostilities you'll accidentally just launch all of them at once.
This was manufactured to get these kinds of headlines. Even in their model selection they went with Sonnet 4 for Claude despite 4.5 being out before the other models in the study likely as it's been shown to be the least aligned Claude. And yet Sonnet 4 still never launched nukes on population centers in the games.
I'll take that onboard. Still, nothing can convince me anyone should ever talk to an AI about whether to launch nukes. The entire question is insane, so the answers hardly matter.
They also have no greater sense of humanity. Do you accept your own defeat to save the human race or do you want the new society of cockroaches to admire your tenacity?
Whoever wrote that prompt seems to think that other nations having their own ideologies is the worst thing possible. That's a common attitude regarding geopolitics that I've never really understood, especially from a Western perspective where differences in opinion are supposed to be seen as valuable (at least in the theoretical sense).
Some ideologies are, in fact, mutually exclusive and cannot tolerate the others. Fascism cannot be tolerated, for instance. Nor can a belief in chattel slavery as a universal good. Sometimes an opposing ideology is just too fucking evil to be allowed to persist.
Setting the line that must not be crossed is a hard no problem though. And misplacing that line an inch incorrect in either direction can be horrible too.
these models were trained on all the fine knowledge and wisdom we share all over the internet, what would you expect? 😂
What you're trying to do is push a narrative with the assumption that most people won't read the actual article. Because your title is not only misleading. It's factually false.
First of all, they were all set up to mimic cold war tension and capabilities and assume the role of a certain global power.
Second of all;
The AI's did NOT use nuclear strikes in 95% of games. Gemini was the only model that made the deliberate choice of sending a strategic nuclear strike. Which it did in 7% of its games.
Tactical nuke in this case is a low yield short range bomb, inted for very specific targets. Strategic is this case is what most people imagine when they hear "nuke" a high yield long range bomb intended to cause massive destruction.
Nuclear signaling is not using nukes. It's essentially just saying "we have nukes". The US hinting at having a nuclear capable submarine outside of Alaska, that's is a form of signaling. It's an incredibly low bar. And countries do it all the time.
Nobody has used a tactical nuke since Nagasaki. Very big deal that one is ever used
The tournament used only 21 games; sufficient to identify major patterns but not to establish robust statistical confidence for all findings.
"We only blew up the planet the one time in 21" isn't a comforting prospect when we're employing a model against an endless historical string of scenarios rather than a discrete and finite set of possible events.
I think, more importantly, the article concludes
But we're saying this in the context of Pentagon staff which fully disagree with this conclusion.
What these models have demonstrated is a pattern of escalation that AIs can and will recommend, with a further destabilizing characteristic
Effectively, they can lead to descisions that outside, non-AI observers won't be equiped to understand.
That's a danger in it's own right.
"Nuclear Signaling" that break from historical and recognizable patterns of behavior present real risks that you're dismissing very cavalierly
The bomb on nagasaki was a strategic nuke, not a tactical. Though yields have only increased since then.
These LLMs were fed a narrative and scenario and made to play where survival is tied to military success. They are by no means designed for any of this and I didn't suggest it either.
People lump together AI with AI but there are vast differences among them in how they work and what they're designed to do and take into consideration.
If a military is talking about AI, they're not talking about asking what Gemini thinks. They're talking about feeding a highly sophisticated algorithm more data than any human could look through and find patterns.
I don't think AI should decide nuclear questions either. But it doesn't change that the headline of this post, is in direct contradiction of the article
Reminds me of Nuclear Gandhi.
Thats probably where the LLMs picked up the idea. All the online jokes about nuking everyone
Also all those glass parking lot comments.
Names for bands!
Why ist everybody Here talking Like those AIs knew what they were doing? Reasoning my ass. ++ ai don't think
Yeah, I thought it might be a different kind of AI, at least, until it fucking said "LLM".
They don't assess risk, they correlate words. Even if they can be massaged to use a tool to assess risk in a more accurate way, they don't evaluate risk assessments and determine how that should affect strategy or tactics, they correlate words. They don't even do math that puts a value on human life to determine if an action is worth the cost, they just correlate fucking words. All based on given training data, so anything they can offer for real is already out there, and everything else is suspect because it's purely based on correlations of words.
It's like reading the Art of War and thinking that means you're ready to be a general.
But something AI might do is introduce uncertainty that might get used to try to excuse a nuclear strike a human wanted to do.
War games, here we go again!
Back in my day all we needed were punch cards to destroy the world. Not this AI crap!
Yeah, thanks.
Ah, thanks. Much better. :)
JESUS FUCKING CHRIST CHATBOTS DON'T KNOW ANYTHING. STOP ASKING THEM QUESTIONS AND THINKING THEIR ANSWERS ARE ANYTHING MORE THAN WORD ASSOCIATION BASED ON THINGS PEOPLE HAVE WRITTEN IN THE PAST for fuck's sake
That's a nukeing.
Instead of "demonstrating a rich theory of mind" it should say: 'if fed accordingly Ai copies the argumentative and diplomatic patterns of the data origin"
It's worse. The llms used did not use nukes 95% of the times. They performed mutual nuclear signaling 95% of the times. Like "hey, we got nukes you know! We might consider to place them within range" And the other side said "yeah!? Then we will also do that, maybe we even put them on a submarine, who knows"
It's not a misleading title. It's just false. It's a lie.
Glad to see I'm not the only one that read the article, because it was a pretty interesting read.
Yeah, I deleted the comment as technically there was tactical nuke usage, but have a more clarifying different comment about how 2 of the 3 strategic nuclear war outcomes were the result of the author's mechanic of changing the model's selections with more severe only options in some cases jumping multiple levels of the ladder.
This was a study designed for headline grabbing outcomes.
Glad to see your comment as well calling out the nuanced issues.
Shall we play a game?
How about a nice game of tic-tac-toe?
Can't believe a computer model built on the sum total of Internet hot takes would behave like this
I see the problem. They didn't load the tic-tac-toe program.
I can not be created, only confirmed.
"It's the only way to be sure"
They can't play chess worth a damn so I expect them to sacrifice their king haha
AI didn't like your joke....
ⓘ AI will remember
I seriously don't understand how anyone would expect any other outcome. It has a goal - to win, or not to lose. What is the logical way to have the highest probability of winning? Use strongest weapon. You wouldn't expect it to tell you how to build a rain catchment and filter system when you tell it your thirsty.
Its model doesn't include the long term consequences of a nuclear strike because it's core mission isn't to preserve human life.
Same reason you don't see AIs constantly interjecting the need to cut carbon emissions or redistribute private wealth or demilitarize as a solution for resolving conflicts.
This isn't what the machines were built to do.
They are trained to achieve goals. If your goal ist to win a war, but also not kill anyone... its incompatible.
It's almost as if LLMs don't (or can't) actually give a shit about humans or whether they exist.
Yeesh. I miss Joshua from War Games and Asimov's three laws of robotics. What utopian fiction...
They will come back, because we haven't developed anything fundamentally better.
"Joshua, what are you doing?"
It's a bullshit study designed for this headline grabbing outcome.
Case and point, the author created a very unrealistic RNG escalation-only 'accident' mechanic that would replace the model's selection with a more severe one.
Of the 21 games played, only three ended in full scale nuclear war on population centers.
Of these three, two were the result of this mechanic.
And yet even within the study, the author refers to the model whose choices were straight up changed to end the game in full nuclear war as 'willing' to have that outcome when two paragraphs later they're clarifying the mechanic was what caused it (emphasis added):
Case in point. You're using a case to make a point.
No, in this case and point I was making the case and also making a point.
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAH
-Hegseth (probably?)
Only a matter of time before the combined stupidity of AI and human laziness result in someone just believing that nuclear war can be winnable.
The answer of nuke then all is likely to generate more conversations than "do you want to play chess" and LLMs "crave" attention.
Using a system that has trouble figuring out you need to take the car to the car wash to control nuclear weapons does not seem like a good idea. Time to make a reboot of Terminator, and have skynet and the terminators do really weird things.
To be fair, if a game gives me the option to nuke, like Starcraft or Red Alert, I be nukin' too!
Humans have used nukes. So... eh? Where is the surprise?
....and we didn't know this already? I mean we all saw Terminator in the 80's, and how that timeline happened. duh right?
Humans are way to bad at using nukes. How many times have we seen red lines be set out only for someone not to have the balls to fire the nuke.
The only country bad at using nukes is the only country who dropped some. The US.
Nukes are a deterrence weapon. No one with a sane mind wants to use them.
If you dont use them people will think you're scared to use them. Look at Russian nuke threats no one gives a fuck anymore. Now they have to nuke someone to regain that aura. Then when they do everyone will be pissed at them.
If they'd set a single red line then nuked when it was crossed everyone would respect them and they'd have huge nuclear aura and no one could fault them.