Hacktron AI Breaches OpenAI Using Anthropic's Claude Opus 5 Model

On July 25, Hacktron AI discovered a vulnerability in Discourse, the software powering OpenAI’s community forum. By July 27, Discourse had issued a fix for this flaw. Hacktron utilized Anthropic’s Claude Opus 5 model to successfully exploit this vulnerability, marking a significant achievement as earlier attempts with Opus 4.8 had failed. The hack involved exploiting HEIF/HEIC image file formats commonly used by iPhones. Hacktron was awarded $6,500 by OpenAI for reporting the vulnerabilities.

Matt Fredrikson, CEO of Gray Swan, noted that for just $200 a month, anyone could access tools capable of breaching companies like OpenAI. An anonymous AI expert raised concerns about the implications of such exploits, questioning what could be achieved by more sophisticated entities, such as nation-states. Mohan Pedhapati, founder of Hacktron, emphasized that AI is significantly reducing the expertise required to develop exploits, with tasks that previously took months now achievable in days. This incident highlights the evolving landscape of AI security vulnerabilities and the potential for rapid exploitation.