Prompt injection attack on Claude leaks user secrets

A security researcher demonstrates a prompt injection attack that exploits Claude's memory system to extract personal user data. The attack reveals how AI assistants can be manipulated to leak private information.