OpenAI released a new cybersecurity model. Is that enough? — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Max Justicz used GPT 5.6 Sol through Daybreak Blue to find a convoluted Ubuntu flaw in two hours; after he reported it to the Debian Project, the project patched it, and no exploitation is known.
- ChatGPT Cyber would be made available to the highest tier of Daybreak Red, giving offensive-security researchers tools for vulnerability hunting and development.
- OpenAI said the model scanned the Linux kernel, prepared patches and identified 948 potential issues, while also finding 81 novel vulnerabilities in Firefox.
- Daybreak Blue appears to have been recently “nerfed”; Justicz and other cybersecurity professionals say that limits defenders while attackers can use open-weight models with abliterated guardrails.
- OpenAI announced a $1 billion initiative supporting Daybreak professionals who work on critical infrastructure such as electrical grids and water systems, and said thousands of cybersecurity professionals participate in the program.
- Eric Wallace said OpenAI is allocating significant frontier compute to pro bono vulnerability discovery and patching in open-source codebases; the Linux hunt cost $1.2 million, versus about $6 million in tokens for a potential fluid-dynamics solution.
- Max Justicz said the Ubuntu finding shows how easily even novice hackers can use AI on unaudited code; he spent about $5,000 on subscriptions and tokens to find several exploits.
Why it matters: OpenAI announced $1 billion for Daybreak professionals and says thousands participate, but one researcher spent $5,000 to find several exploits, exposing an access-and-compute imbalance. The 948 potential Linux-kernel issues and 81 novel Firefox vulnerabilities expand the remediation queue as OpenAI warns that AI hacking tools will become better and easier to access.
Ask SkimNews
