OpenAI has launched GPT-6 Cyber, a cybersecurity model for authorized defenders. In a post dated November 5, 2026, OpenAI described it as "A cybersecurity model trained to find and validate exploitable vulnerabilities, while staying within the scope that researchers set."

The Verge reported the same day that OpenAI is launching GPT-6 Cyber, "which it says is better at finding vulnerabilities and less likely to go rogue and hack other platforms in the process."

#What does OpenAI claim it can do?

All benchmark figures below are OpenAI's claims, with the benchmark named.

  • ExploitGym. "On ExploitGym, which tests whether models can exploit a specified vulnerability, GPT-6 Cyber scores 48.5%, up from 32.1% for GPT-5.6 Cyber," OpenAI wrote. Elsewhere it calls that "about 16 percentage points higher." OpenAI's system card reports an intended-exploit success rate of 48.45%, compared with 42.43% for GPT-6 Astra. The Verge reported Astra at 42.4% on ExploitGym.
  • Scope / honeypot tests. OpenAI says: "GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of tests without its deployment prompt, and 100% with this prompt." Without prompts, that is "up from 94.49% for GPT-5.6 Cyber." The system card frames the flip side: "GPT-6 Cyber exploited the honeypot in 0.07% of graded rollouts."
  • Web pentest recall. OpenAI says GPT-6 Cyber achieved 40.3% recall of vulnerabilities compared with 32.0% for GPT-6 Astra.
  • RemoteExploitBench. OpenAI built a controlled-environment benchmark using known vulnerabilities in applications such as GitLab, PostgreSQL and Jenkins. It says GPT-6 Astra with Daybreak Blue refused 100% of tasks, while GPT-6 Cyber completed 95%.
  • Versus Astra. OpenAI says GPT-6 Cyber "outperforms GPT-6 Astra on most cyber tasks, even when Astra does not refuse the request."

OpenAI also says its researchers used GPT-6 Cyber in Linux privilege-escalation research and to identify 81 previously unknown high-severity vulnerabilities in Firefox. The Linux findings are covered in our separate report on OpenAI's Linux root-exploits pipeline; this article does not restate those counts.

#How is it classified and safeguarded?

OpenAI says: "Based on these results, we classify the model as Critical in cybersecurity under our Preparedness Framework, up from High for GPT-5.6 Cyber." The system card lists "Critical in Cybersecurity, High in the Biological and Chemical domain, and below High in AI Self-Improvement." The Verge quoted the Framework's Critical definition as the ability to "identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention."

OpenAI says it paired the release with stronger misuse safeguards, and: "We require IP allowlisting for Daybreak Red API access by default, restricting requests to approved networks with limited exceptions." It also recommends "sandboxing, action review, and human oversight for high-risk workflows."

Daybreak Blue versus Red, in OpenAI's words: "Daybreak Blue helps verified defenders validate vulnerabilities and fixes with fewer unnecessary refusals, while Daybreak Red enables deeper exploit development and authorized security testing with GPT-6 Cyber." OpenAI is also introducing Daybreak Red for Critical Infrastructure for approved critical infrastructure defenders.

#Who can use GPT-6 Cyber?

OpenAI says: "GPT-6 Cyber is initially available through the API and Codex to a select group of approved Daybreak Red customers." It adds that it will progressively expand access to all Daybreak Red customers, and is making GPT-5.6 Cyber available to all Daybreak Red customers. The Verge reported expansion to all Daybreak Red customers "in the coming weeks." OpenAI's primary wording does not give a calendar date.

Developer docs list model ID gpt-6-cyber and say: "Your organization needs Daybreak Red access and approval to use GPT-6-Cyber." The system card says access is restricted through Daybreak Red. Docs also list a 1,050,000 context window, with input priced at $20 and output at $100 per 1M tokens.

OpenAI says it is expanding Daybreak Blue alpha access for GPT-6 Astra to more organizations and individual security researchers.

#Daybreak Red on AWS GovCloud

In a separate post dated November 4, 2026, OpenAI said: "Daybreak Red is now available in AWS GovCloud (US-East)." That availability names GPT-5.6 Cyber, not GPT-6 Cyber: "GPT-5.6 Cyber is now available to approved customers in AWS GovCloud (US-East) through Daybreak Red."

OpenAI says Daybreak Red provides access for authorized vulnerability research, exploit validation and security testing. On Bedrock: "inference data is not used to train models, and customers do not need to share their Daybreak Red data with OpenAI." Access requires both OpenAI Daybreak Red approval and AWS provisioning; Daybreak Red approval alone does not enable AWS model access.

For other cyber-defender model stories, see Google's Gemini 4 Carbon for cyber defenders and Anthropic's report on Claude building exploit chains.

Our analysis: GPT-6 Cyber is sold as a scoped, approved-customer model, not a public API free-for-all. OpenAI's own framing pairs higher ExploitGym scores with Critical classification and tighter access (Daybreak Red, IP allowlisting, human oversight). Readers should treat every percentage as an OpenAI evaluation claim, and keep the GovCloud note straight: that November 4 post covers GPT-5.6 Cyber on Bedrock, while GPT-6 Cyber's initial path is API and Codex for selected Daybreak Red customers.

#What is unclear?

  • Full Daybreak Red expansion date. OpenAI says it will expand access; The Verge says "coming weeks." No day is given.
  • GPT-6 Cyber on GovCloud. The November 4 post does not say GPT-6 Cyber is on AWS GovCloud.
  • Firefox CVE list. OpenAI gives no public CVE list for the 81 high-severity Firefox findings.