Founder tested LLM security exploits on his own app for $1,500

jc4p deliberately built vulnerabilities into an app and hired LLMs to find them, documenting what worked and what didn't. The experiment reveals both the potential and limitations of AI-assisted penetration testing—useful for spotting certain classes of bugs, but nowhere near a substitute for human security review.