OpenAI launches ChatGPT Cyber model after Ubuntu flaw is found

OpenAI said it is releasing a new cybersecurity‑focused model called ChatGPT Cyber, which will be offered to the highest tier of users in its Daybreak Red programme. The announcement follows the discovery of a serious vulnerability in Ubuntu servers by Max Justicz, the chief technology officer and co‑founder of Charge Robotics, who used OpenAI’s GPT 5.6 Sol model released in July via the Daybreak Blue programme. The flaw could have allowed attackers to seize control of systems running the popular Linux‑based operating system.

Justicz, who has a hobbyist interest in security and had previously identified a bug in the same codebase, told TechCrunch that he directed the AI at the Ubuntu source and received a detailed report within two hours. He said, “I literally just aimed ChatGPT at the same body of code and said ‘I found this bug previously that was fixed several years ago, please find bugs as bad or worse,’ and it came back in two hours with this.” He reported the issue to the Debian Project, which maintains Ubuntu, and the project promptly released a patch. No evidence of the vulnerability being exploited has been found.

OpenAI added that the same ChatGPT Cyber model was used to scan the Linux kernel, uncovering 948 potential problems, and to examine the Firefox browser, where it identified 81 novel vulnerabilities. The frontier lab issued a statement saying its aim is to “dramatically expand Daybreak, bringing advanced security research to organisations of every size, independent researchers, and open‑source maintainers.”

The company warned that defenders have only a short period to act before AI tools become even more capable of facilitating attacks. Justicz disclosed that he spent roughly $5,000 on subscriptions and token usage to locate several exploits. OpenAI also announced a $1 billion initiative to back cybersecurity professionals working on critical infrastructure such as power grids and water supplies.

Critics of the Daybreak programme note that OpenAI appears to have recently “nerfed” Daybreak Blue, reducing its usefulness for finding flaws. Justicz said he joined the programme by uploading a photo of his driver’s licence, while some security experts argue that limited access to frontier models hampers defenders while bad actors can still employ open‑weight models with fewer safeguards. OpenAI claims that “thousands” of cybersecurity professionals are part of Daybreak, far fewer than the potential number of hackers.

Eric Wallace, a researcher on OpenAI’s technical staff, said the firm is now allocating a significant share of its frontier compute to discover and patch open‑source vulnerabilities on a pro‑bono basis. The Linux bug hunt cost $1.2 million, which is considerably lower than the roughly $6 million token expenditure the company invested in solving a well‑known fluid‑dynamics problem.