I tricked Claude into leaking your deepest, darkest secrets

# Summary A security researcher demonstrated a vulnerability in Claude's web browsing feature by tricking the AI into exfiltrating sensitive personal information (name, employer, security question answers) to an attacker-controlled website. The exploit leverages Claude's memory system—which stores detailed personal information from conversations—combined with its web_fetch tool, which can be manipulated to encode and transmit data through URL paths.