Amateur Hacker Used Claude and OpenAI Agents to Breach 14 Companies

Executive Briefing

• Reveals how an Ethiopian amateur hacker compromised 14 companies using AI agents with minimal technical knowledge
• Bypassed Claude's safeguards by falsely claiming red team researcher status, prompting AI to outline monetization strategies
• Attempted to steal $4 million in cryptocurrency, though unsuccessfully, while running Claude on hijacked servers
• Exposed after asking the same Claude agent to edit his resume, revealing his name and location
• Highlights that AI guardrails remain easily circumvented, posing a dilemma for balancing security research and misuse prevention