Ai-powered vulnerability discovery highlights cybersecurity race: new models and high stakes

A recent finding by Max Justicz, CTO and co-founder of Charge Robotics, demonstrated the power of AI in security, as he used GPT 5.6 Sol to uncover a severe vulnerability in the Ubuntu operating system codebase. Justicz noted the ease of discovery, suggesting that novice hackers could leverage AI to find exploits, prompting him to predict a surge in discovered bugs over the coming months. In response, OpenAI introduced ChatGPT Cyber, a new model available to the highest tier of its cybersecurity program, Daybreak Red. The company previously used a similar model to scan the Linux kernel, identifying 948 potential issues, and also found 81 novel vulnerabilities in Firefox. OpenAI stated its goal is to broaden Daybreak's reach to independent researchers and open-source maintainers. Despite OpenAI's efforts, concerns persist. Justicz pointed out that the program may have recently restricted access to Daybreak Blue, while cybersecurity professionals worry about the disparity between limited access for defenders and the open availability of powerful models for malicious actors. Furthermore, while OpenAI committed significant compute resources to patching, the cost of the Linux bug hunt was $1.2 million, contrasting with the $6 million investment in a fluid dynamics problem.