Discover how a new exploit called Policy Puppetry can extract hidden system instructions from any major AI model, including ChatGPT, Claude, Grok, and Gemini. Learn how it works, why it matters, and what it reveals about AI security.