A new jailbreak called "Policy Puppetry" can bypass safety guardrails on every major AI model, including ChatGPT, Claude, Gemini, and Llama, using a single prompt. Discover how it works, why it matters, and what this means for the future of AI safety.