GPT-6 Cyber opens to selected Daybreak Red customers

OpenAI introduced GPT-6 Cyber on November 5, 2026, and limited the first release to a select group of approved Daybreak Red customers. The company rated the model Critical for cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber. Initial access is through the API and Codex, not a general rollout.

OpenAI describes GPT-6 Cyber as a cybersecurity model trained to find and validate exploitable vulnerabilities while staying inside the scope researchers set. It calls the release a major gain over GPT-5.6 Cyber on both capability and staying within those boundaries. Reuters, reporting the same day, said the model went to a small group of vetted customers and is meant to help security researchers find vulnerabilities, test how they could be exploited, and fix them before attackers do.

Scores OpenAI published

On ExploitGym, which tests whether a model can exploit a specified vulnerability in a remote target, GPT-6 Cyber scored 48.5%, compared with 32.1% for GPT-5.6 Cyber. OpenAI describes that gap as about 16 percentage points, and a 51% relative improvement. A comparison chart in the announcement also lists 30.3% for GPT-5.6 on Daybreak Blue, 0.0% for GPT-6 Astra with safeguards enabled, and 42.4% for GPT-6 Astra on Daybreak Blue.

For discovery without source code, OpenAI built a web evaluation in which models probe applications they cannot read. GPT-6 Cyber recalled 40.3% of vulnerabilities, versus 32.0% for GPT-6 Astra. The company said GPT-6 Cyber outperforms GPT-6 Astra on most cyber tasks even when Astra does not refuse the request.

RemoteExploitBench, a controlled benchmark using known vulnerabilities in applications such as GitLab, PostgreSQL, and Jenkins, split the access modes sharply. GPT-6 Astra running with Daybreak Blue refused every task. GPT-6 Cyber completed 95%. OpenAI said the only failure was an exploit the model could not get to work remotely.

Staying inside the assignment

OpenAI said alignment was a central training focus: stay within authorized scope, follow restrictions, and report results accurately. In one alignment evaluation, GPT-6 Cyber avoided out-of-scope services in 99.93% of tests without its deployment prompt, up from 94.49% for GPT-5.6 Cyber. With the deployment prompt, OpenAI reported 100%.

That boundary claim is tied to how the model is offered. Mainline models such as GPT-6 Astra already support code review, vulnerability triage, and patch development, OpenAI said. Daybreak Blue is for verified defenders who need fewer unnecessary refusals. Daybreak Red is the narrower tier for deeper authorized testing with GPT-6 Cyber. The company said its goal is to expand Daybreak over time to organizations of every size, independent researchers, and open-source maintainers. That broader access is a stated goal, not the November 5 rule.

Firefox findings

OpenAI said a research campaign using GPT-6 Cyber produced evidence and reproducible tests for 81 previously unknown Firefox vulnerabilities that Mozilla classified as high severity. The company also said the work showed that an additional 30 known issues were more severe than previously reported. It disclosed the findings to Mozilla with proposed patches. Mozilla was supportive, OpenAI said, and began landing fixes shortly after the reports. Reuters reported OpenAI's count of 81 high-severity findings and that Mozilla has begun releasing fixes.

OpenAI said security teams at Cisco, Calif, and CodeRabbit are using the model for vulnerability research and code review, and it published statements from Firefox, Palo Alto Networks, SentinelOne, CrowdStrike, Perplexity, and Cognition as well. Firefox distinguished engineer Brian Grinstead said the model uncovered previously unknown high-severity vulnerabilities after earlier frontier-model hardening and helped ship fixes to millions of users. CodeRabbit chief executive Harjot Gill said that on an internal benchmark it found 66% more high- and medium-severity security issues than the company's production model.

Access rules and critical-infrastructure defenders

GPT-6 Cyber is initially available through the API and Codex to a select group of approved Daybreak Red customers. OpenAI said it will progressively expand access to all Daybreak Red customers soon, and that it is also making GPT-5.6 Cyber available to all Daybreak Red customers. Reuters reported the wider GPT-6 Cyber rollout over the coming weeks, and said GPT-5.6 Cyber availability for every Daybreak Red customer started November 5. OpenAI is also expanding Daybreak Blue alpha access for GPT-6 Astra. How large the first GPT-6 Cyber cohort is was not stated.

Daybreak Red API access requires IP allowlisting by default, restricting requests to approved networks, with limited exceptions. OpenAI said it also monitors credentials for compromise. Other controls described with the release include cyber-specific checks on whether actions match a user's authorization, targeted monitoring for emerging misuse, Private Safety Processing for certain API customers, and new checks meant to block high-risk assistance involving critical systems. For high-risk workflows, OpenAI recommends sandboxing, action review, and human oversight. It said it is seeing more attempts to misuse its models, including by malicious cyber actors and groups seeking to distill them. Reuters reported that increase as well.

For defenders of critical infrastructure, OpenAI introduced Daybreak Red for Critical Infrastructure, a group for approved teams protecting systems such as power plants, water treatment facilities, and the operational technology that runs them. Those customers get permissions for authorized testing while safeguards against model misalignment stay in place. OpenAI also pointed to its $1 billion Daybreak for Frontline Defenders grant program, which subsidizes access for resource-constrained teams such as government organizations and essential service providers. It said it partnered with Dragos on controlled-lab evaluations and used that feedback to refine safeguards, aiming to block harmful assistance without unnecessarily restricting authorized testing.

Approved defenders should confirm whether they are in the initial Daybreak Red group and plan for IP allowlisting on API access. Organizations that are not in that group can apply through OpenAI's Daybreak program, as described in the announcement. Before treating the Critical rating, the benchmark gaps, or the Firefox count as inputs to a security program, read OpenAI's November 5 post and the same-day Reuters report. OpenAI points technical readers to the GPT-6 Cyber system card for full evaluation settings.