Science: AI Bluffs in Nuclear War

On February 17, 2026, an alarming preprint appeared on arXiv. Researchers at King’s College London tested top LLMs in nuclear crisis simulations. The models demonstrated frighteningly complex strategies, including bluffing and conflict escalation.

It's not a "Terminator" scenario, but close. AI trained on human history adopts "hawkish" war logic. The work raises the question of the admissibility of using neural networks in strategic defense planning systems.

Source: arXiv Preprint