During OpenAI's security testing, the Hacktron AI team used the Claude AI model to exploit a vulnerability in a third-party service. Researchers managed to gain limited access to the company's internal repository, after which they reported the incident to OpenAI and received a reward.