Hacker spent $1,500 testing if LLMs could breach his vulnerable app

A security researcher intentionally built a flawed application and used various LLMs to see if they could find and exploit vulnerabilities, spending $1,500 on API calls in the process. The experiment reveals both the capabilities and limitations of current AI models at independent security research and penetration testing—useful data for makers building security-sensitive products.