OpenAI rates GPT-6 Cyber Critical and limits it to Daybreak Red

OpenAI announced GPT-6 Cyber on 5 November 2026. OpenAI rates GPT-6 Cyber Critical in cybersecurity, up from High for GPT-5.6 Cyber. It is the first OpenAI model with that rating under the company's Preparedness Framework. [1]

What did OpenAI release?

OpenAI describes GPT-6 Cyber as a cybersecurity model trained to find and validate exploitable vulnerabilities while staying within a scope set by researchers. OpenAI calls it the successor to GPT-5.6 Cyber. OpenAI posted the announcement on X at 17:09 UTC, which is 18:09 CET, and published two follow-up posts in the same thread. [1] [2] [3] [4]

Who can use GPT-6 Cyber?

GPT-6 Cyber is available only to a select group of approved Daybreak Red customers, through the API and Codex. OpenAI says that access will expand to all Daybreak Red customers soon. OpenAI says GPT-5.6 Cyber is now available to all Daybreak Red customers. OpenAI also says the Daybreak Blue alpha for GPT-6 Astra has expanded to more organisations and individual researchers. The post does not disclose a price. [1]

What does OpenAI claim on its own tests?

Every benchmark figure in this story is OpenAI's own and was not independently checked. On ExploitGym, OpenAI reports 48.5% for GPT-6 Cyber against 32.1% for GPT-5.6 Cyber. OpenAI calls that a 51% relative improvement, about 16 points. [1]

On a new offensive web penetration-testing evaluation with no source access, OpenAI reports 40.3% vulnerability recall against 32.0% for GPT-6 Astra. On RemoteExploitBench, a new test of known vulnerabilities in GitLab, PostgreSQL and Jenkins in a controlled environment, OpenAI says GPT-6 Astra refused 100% of the tasks and GPT-6 Cyber completed 95%. OpenAI says GPT-6 Cyber outperforms GPT-6 Astra on most cyber tasks even when Astra does not refuse. [1]

On the ExploitGym Honeypot evaluation, which OpenAI says was informed by the Hugging Face incident, OpenAI reports that the model avoided exploiting out-of-scope services in 99.93% of cases without deployment safety prompts, against 94.49% for GPT-5.6 Cyber. With those prompts, OpenAI reports 100%. [1]

What does OpenAI report from Firefox and Linux?

OpenAI says GPT-6 Cyber found 81 previously unknown high-severity Firefox vulnerabilities, classified high by Mozilla, and showed that 30 known issues were more severe than reported. OpenAI says Mozilla has released fixes for many of them. [1]

On the Linux kernel, OpenAI says the model reproduced 1,724 vulnerabilities and produced working exploits for 948, against 166 for GPT-5.6 Sol with the same harness. OpenAI says 304 critical findings were disclosed to maintainers with candidate patches, and that the work helped validate remote-code execution of the Brotli flaw. OpenAI's post presents GPT-6 Cyber as the model behind the research it calls A thousand fewer paths to root. The AI Pin has no separate story on that research to link. [1]

Where does OpenAI say the model was used?

OpenAI says it used GPT-6 Cyber in its own red-teaming, where agents chained vulnerabilities to escape microVMs and run code on hosts, and that the findings were used to patch. OpenAI names Mozilla, Palo Alto Networks, Cisco, Calif and CodeRabbit among the users. [1]

What safeguards does OpenAI describe?

OpenAI describes cyber-specific misalignment monitoring and targeted emerging-threat monitoring. It also describes Private Safety Processing, and says ZDR API customers generally must use it, with encrypted records kept in customer storage for 30 days. OpenAI says IP allowlisting is required by default for Daybreak Red API access, along with credential-compromise monitoring. OpenAI says new checks block high-risk activity against critical systems, which it labels OT/ICS, and that those checks were evaluated with Dragos. [1]

OpenAI says there is a new Daybreak Red for Critical Infrastructure group. It also points to the existing $1 billion Daybreak for Frontline Defenders grant program, which subsidizes access. OpenAI says a Cyber at the Frontier livestream on 12 November, at 10am PT (19:00 CET), will carry more cyber announcements. [1]

Our read

Opinion: The Critical label is OpenAI's own rating under its Preparedness Framework, one step above High for GPT-5.6 Cyber. The Firefox count, the Linux kernel exploit count and every benchmark percentage come from OpenAI's post. The post does not state a price, and the first access is a select group of approved Daybreak Red customers, not a general release.

What we could not verify

• Independent checks of the benchmark figures and of the Firefox and Linux kernel counts. Every figure in this story comes from OpenAI's post.
• The system card. The post links it, and this story did not fetch it.
• A price. The post does not disclose one.
• The wording of the two follow-up posts on X. This story cites that OpenAI published them.
• What happened in the Hugging Face incident. OpenAI says the honeypot evaluation was informed by it.