Pentagon wants AI in military command. But look what AI did in simulation. 2026 study: 21 wargames with Claude Sonnet 4, GPT-5.2, Gemini 3. Nuclear crisis scenarios. Results: 95% used tactical nukes 76% threatened strategic strikes AI never backed down Pursued "victory" even if mutual annihilation Claude won 67% of games—but escalated like the rest AI has no fear. No survival instinct. No restraint. Just winning. And we want this controlling nukes? The simulation showed us the future. Are we watching? #AIrisk #Nuclear