OpenAI launches GPT-6 Cyber for security researchers with critical rating
The new model aims to help researchers identify and fix security vulnerabilities before malicious actors can exploit them.
OpenAI released GPT-6 Cyber to a small group of vetted customers on Thursday, according to Firstpost. The model is designed to help security researchers find, test, and fix vulnerabilities, and has been classified as “critical” under the company's safety framework.
The company reported that researchers used the model to identify 81 previously unknown Firefox vulnerabilities, which Mozilla classified as high severity. Mozilla has started releasing fixes for these flaws. OpenAI also disclosed 304 critical Linux findings and proposed patches to maintainers.
OpenAI stated the launch comes as malicious cyber actors increasingly attempt to misuse AI. The company noted it is seeing attempts by groups to distil its models, which adds to concerns regarding potential misuse.
To prevent misuse, OpenAI introduced several protections, including automated checks and cyber-specific misalignment monitoring. The company also introduced safeguards for Daybreak Red API access, such as default IP allowlisting and monitoring credentials for potential compromise.
OpenAI partnered with Dragos to test safeguards and refine the model to block harmful assistance without restricting authorised testing. The company has also increased investment in threat investigations involving operational technology and industrial control systems.
GPT-6 Cyber is initially available through the API and Codex to select Daybreak Red customers. Access is expected to expand to all approved customers over the coming weeks, while the previous GPT-5.6 Cyber model is being made available to all Daybreak Red customers starting Thursday.
