A universal jailbreak called Policy Puppetry is exposing deep flaws in AI safety. Learn how simple prompt tricks can bypass guardrails on ChatGPT, Claude, Gemini, and more, and what it means for the future of AI alignment.