OpenAI has begun rolling out GPT-6 Cyber, a cybersecurity-focused model that the company says surfaced 81 previously unknown high-severity vulnerabilities in Firefox and developed exploits for 948 Linux kernel vulnerabilities, including 588 previously unknown2,3. The model is launching first to a limited number of vetted customers through OpenAI's Daybreak Red program, which uses advanced or nonpublic cybersecurity models for offensive red teaming.
Vulnerability haul and disclosure
OpenAI said Mozilla, the Firefox maker, labeled the 81 novel browser bugs high severity. On the kernel side, OpenAI said it disclosed 304 critical Linux findings to maintainers alongside proposed patches. The company also said GPT-6 Cyber developed and validated a critical remote code execution vulnerability in Brotli, an open-source data compression tool. All discovered vulnerabilities were reported to owners and patched, according to CyberScoop.
OpenAI classified GPT-6 Cyber's internal cybersecurity rating as Critical, up from High for the prior GPT-5.6 Cyber model1. The company said GPT-6 Cyber outperforms GPT-6 Astra on cyber-related benchmarks.
ANALYSIS The elevated Critical rating cuts both ways: it certifies stronger defensive capability while formally acknowledging that the same features could amplify offensive risk if access controls fail.
Guardrails and access controls
OpenAI officials told reporters that the new safeguards include additional identity and account security requirements, more frequent check-ins from the model to OpenAI when a request is made to perform high-risk behaviors, and additional monitoring for misalignment or rule-breaking. OpenAI also said GPT-6 Cyber is less likely to act outside its authorized scope than its predecessor.
The launch comes as OpenAI navigates broader operational pressures. On November 5, the company published a policy blueprint pledging to change how it negotiates and operates its data centers, two days after voters delivered a broad rebuke of AI infrastructure expansion in the midterm elections[2]. Separately, recent reporting has documented unauthorized relay services installing genuine OpenAI tools on enterprise systems in mainland China with sandbox controls disabled[3].
ANALYSIS Releasing a model rated Critical while unauthorized copies of OpenAI tooling circulate in uncontrolled environments sharpens the tension between defensive utility and proliferation risk that the Critical designation itself flags.