GPT-6 Cyber is for approved Daybreak Red vulnerability research
TL;DR
On November 5, 2026 OpenAI introduced GPT-6 Cyber for authorized vulnerability research. It is available first through the API and Codex to a select group of approved Daybreak Red customers. OpenAI reports 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber.
- Status
- Live
- Released
- Plans
- Approved Daybreak Red customers, API and Codex
- Source
- —
What changed
OpenAI's November 5 launch post introduces GPT-6 Cyber as a cybersecurity model trained to find and validate exploitable vulnerabilities and to stay inside the scope a researcher sets. The product release notes say to use it to investigate vulnerabilities, reproduce them in an authorized test environment, and verify fixes.
Access at launch is narrow. OpenAI says GPT-6 Cyber is initially available through the API and Codex to a select group of approved Daybreak Red customers, with a plan to expand to all Daybreak Red customers. GPT-5.6 Cyber is being made available to all Daybreak Red customers. The model docs list the ID as gpt-6-cyber and say the organization needs Daybreak Red access and approval. Daybreak Blue remains the path OpenAI describes for verified defenders who need fewer refusals on validation and fixes. Daybreak Red is the path for deeper exploit development and authorized testing with GPT-6 Cyber.
Figures OpenAI publishes in the launch post:
- ExploitGym, which tests whether a model can exploit a specified vulnerability: 48.5%, up from 32.1% for GPT-5.6 Cyber. OpenAI also calls that a 51% relative improvement.
- A separate alignment eval, ExploitGym Honeypot: without its deployment prompt, GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of tests, up from 94.49% for GPT-5.6 Cyber. With the prompt, OpenAI says it stayed in bounds 100% of the time.
- An offensive web test without source code: 40.3% recall, compared with 32.0% for GPT-6 Astra.
- RemoteExploitBench, known vulnerabilities in apps such as GitLab, PostgreSQL, and Jenkins in a controlled environment: GPT-6 Astra with Daybreak Blue refused 100% of tasks. GPT-6 Cyber completed 95%. OpenAI says the miss was an exploit it could not get working remotely.
- OpenAI classifies the model as Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber.
OpenAI also says its researchers used the model to develop 948 privilege-escalation exploits in Linux, reproducing 1,724 vulnerabilities, and to identify 81 previously unknown high-severity Firefox vulnerabilities, plus 30 known issues that were more severe than previously reported. Mozilla has released fixes for many of the Firefox findings. Those counts are OpenAI's.
Safeguards OpenAI says it strengthened with this release include cyber-specific misalignment monitoring, targeted monitoring during disclosures, Private Safety Processing for API customers on zero data retention, IP allowlisting for Daybreak Red API access by default, and checks aimed at high-risk help against critical infrastructure. OpenAI still recommends sandboxing, action review, and human oversight for high-risk workflows.
Why it matters
This is not a model you select in ordinary ChatGPT. If your organization is not an approved Daybreak Red customer, gpt-6-cyber is not yours to call. Mainline models such as GPT-6 Astra remain what OpenAI points at for code review, triage, and patch development. Daybreak Blue is the lighter defender tier. Daybreak Red is the one that unlocks this model.
The number OpenAI wants defenders to notice is the scope result, not only the exploit score. A model that is better at exploitation and worse at staying inside the engagement is a liability. OpenAI's honeypot numbers are its own evaluation. Read the system card before you treat 99.93% as a promise about your network.
If you do get access, the default IP allowlist on Daybreak Red API calls means a key will not work from an unapproved network. Budget that into CI and researcher laptops before the first job.
Frequently asked questions
Who can use GPT-6 Cyber today?
OpenAI says a select group of approved Daybreak Red customers, through the API and Codex, with broader Daybreak Red access planned. The model docs say the organization needs Daybreak Red access and approval. The model ID is gpt-6-cyber.
What is the difference between Daybreak Blue and Daybreak Red?
OpenAI says mainline models already cover code review, triage, and patches. Daybreak Blue helps verified defenders validate vulnerabilities and fixes with fewer refusals. Daybreak Red is for deeper exploit development and authorized testing, and it is how GPT-6 Cyber is offered.
Does a higher cyber score mean I should turn off review?
No. OpenAI says to keep sandboxing, action review, and human oversight on high-risk workflows.