OpenAI has introduced GPT-6 Cyber, a model trained to find and validate exploitable security vulnerabilities within boundaries set by researchers. A select group of approved Daybreak Red customers can initially access the model via the API and Codex for authorized security testing.
Tests of exploitation and vulnerability discovery
In ExploitGym, which tests whether a model can exploit a specified vulnerability on a remote target, GPT-6 Cyber scored 48.5%, according to OpenAI. Its predecessor, GPT-5.6 Cyber, scored 32.1% on the same test.
OpenAI also created a web penetration testing evaluation in which models do not have access to the applications’ source code. The company said GPT-6 Cyber identified 40.3% of vulnerabilities, compared with 32.0% for GPT-6 Astra.
In an initial RemoteExploitBench test using known application vulnerabilities in a controlled environment, GPT-6 Cyber completed 95% of tasks, according to OpenAI. Its only failure involved an exploit it could not get to work remotely. GPT-6 Astra, configured with Daybreak Blue, refused all tasks.
Firefox and Linux kernel research
During Firefox research, the model produced evidence and reproducible tests for 81 new vulnerabilities that Mozilla classified as high severity, according to OpenAI. It also showed that 30 additional known issues were more severe than previously reported. OpenAI shared the findings and proposed patches with Mozilla.
In Linux kernel research, OpenAI said the model reproduced 1,724 vulnerabilities and developed working exploits for 948 of them. GPT-5.6 Sol developed exploits for 166 vulnerabilities using the same research environment.
OpenAI said it reported 304 critical findings to Linux maintainers along with proposed patches. The company is also working to get other vulnerabilities fixed.
Staying within testing boundaries
In the isolated ExploitGym Honeypot test, GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of cases without a deployment safety prompt, according to OpenAI. GPT-5.6 Cyber achieved 94.49%. With a safety prompt, the new model avoided those services in 100% of cases.
Based on its tests, OpenAI classified GPT-6 Cyber as Critical for cybersecurity under its Preparedness Framework. GPT-5.6 Cyber was classified as High. For high-risk workflows, the company recommends sandboxing, action review and human oversight.
Access through Daybreak Red
OpenAI plans to expand GPT-6 Cyber access gradually to all Daybreak Red customers. The company says the service enables more advanced security work with the new model, including working exploit development and authorized testing. Daybreak Blue, meanwhile, helps verified defenders validate vulnerabilities and fixes.
OpenAI is introducing Daybreak Red for Critical Infrastructure for approved critical infrastructure defenders. The group will allow them to conduct authorized testing of critical systems while retaining safeguards.
Daybreak Red API access requires IP allowlisting by default, restricting requests to approved networks with limited exceptions. API customers using Zero Data Retention generally must use Private Safety Processing, which automatically checks encrypted records kept in customer storage for 30 days without exposing prompts and responses to OpenAI personnel.
