Researchers Breach OpenAI Systems Using Anthropic's Claude Model

Researchers at cybersecurity firm Hacktron breached parts of OpenAI's internal systems using Anthropic's latest AI model, highlighting the growing cyber capabilities of advanced AI tools. The team exploited a vulnerability in OpenAI's public help forum and received a $6,500 bounty for reporting the findings. The security test was part of OpenAI's bug hunting programme, and the company has since fixed the vulnerabilities. Hacktron initially struggled with Claude Opus 4.8 but succeeded after Anthropic released Opus 5, showing rapid advances in AI's ability to perform sophisticated technical tasks. The researchers had access to a special version of Claude for qualified cybersecurity practitioners. Hacktron warned that AI could lower the barrier for cyberattacks by making specialised expertise more accessible, potentially reducing work that once required months and well-resourced teams to days. They said security practices would need to adapt as AI models become increasingly capable of assisting with sophisticated cyberattacks.