Cercetători în securitate au demonstrat, folosind Claude, cum pot ajunge la codul intern al OpenAI

Independent security researchers accessed an OpenAI employee's ChatGPT account, exploiting vulnerabilities in the company's systems using AI tools from Anthropic. The Hacktron AI team discovered issues in OpenAI's community forum and authentication system, allowing them to reach internal repositories. They reported the vulnerabilities through OpenAI's bug bounty program, receiving a $6,500 reward. The experiment highlighted the dual-use nature of AI models, which can both identify vulnerabilities and facilitate cyberattacks. OpenAI has since addressed the identified issues and restricted access to affected tokens.