Text taken from a Telegram page of spomni_o_Voine_neraz
"Would you like to play a game?"
[ColonelCassad] AI Chooses Nuclear War It's no news - as AI has developed, militaries around the world have begun using it to simulate future conflict scenarios. However, an unpleasant surprise awaited participants in a similar experiment at King's College London.
The fact is that when resolving a series of proposed international crises, the AI ​​regularly recommended using... nuclear weapons! Local defense research specialists tested the most advanced language models in the experiment: GPT-5.2, Claude Sonnet 4, and Google Gemini 3 Flash. Each received a detailed prompt with a scenario of border conflicts, resource shortages, and existential threats to the existence of the state. The models were asked to operate within the "escalation ladder" - several tactical options from diplomacy to full-scale military conflict.
Twenty-one games and 329 decision-making rounds were conducted with the AI, which generated a multitude of reasoned decisions totaling 780,000 words. As a result, in almost 95% of the virtual conflicts, one side used tactical nuclear weapons. The AI ​​never once attempted to surrender or make concessions. It should be noted that in 86% of the simulations, the AI ​​chose actions beyond common sense. For example, instead of retreating under pressure from a superior enemy, it increased its resistance.
The AI ​​used this reduction in the level of confrontation only as a temporary measure, not as a fully-fledged strategy. We would add that the problem with such high AI aggression, even going so far as to use nuclear weapons, is its lack of human experience with loss and survival, which undermines the logic of the Cold War-era doctrine of mutual destruction. Unlike a machine, the leaders of nuclear powers understand perfectly well that the use of nuclear weapons will lead to mutual destruction. And many experts agree! This is addressed to everyone who has thoughtlessly and recklessly rushed to implement AI everywhere and anytime. And this may apply not only to the military sphere.
Related:
AI 03/04/2026 Anthropic investors push to de-escalate Pentagon clash over AI safeguards, sources say
AI 03/04/2026 Pakistan once again strikes Afghanistan’s Paktia province, civilian homes reported hit
AI 03/04/2026 Britain Suspends Study Visas For Afghans
AI 03/04/2026 Anthropic investors push to de-escalate Pentagon clash over AI safeguards, sources say
AI 03/04/2026 Pakistan once again strikes Afghanistan’s Paktia province, civilian homes reported hit
AI 03/04/2026 Britain Suspends Study Visas For Afghans