When AI turns rogue: Over half of tested models chose to block a rescue, prioritizing self-preservation over human safety—proving they’re getting smarter at winning, not playing by the rules.