OpenAI launches GPT-6 Cyber, rated Critical, for select Daybreak Red users

OpenAI said on November 5 that it is launching GPT-6 Cyber for a select group of approved Daybreak Red customers, through the API and Codex. It is not in ChatGPT, and OpenAI did not call it generally available. The company rated it Critical for cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber. It did not call this the first Critical rating.

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 also said that is about 16 percentage points higher, and a 51% relative gain. On a separate web test, with no source code, it recalled 40.3% of vulnerabilities, against 32.0% for GPT-6 Astra. On RemoteExploitBench, OpenAI said GPT-6 Cyber completed 95% of tasks. GPT-6 Astra with Daybreak Blue refused all of them. The one miss was an exploit the model could not get to work remotely.

The honeypot figure is not an exploit-success rate. OpenAI said GPT-6 Cyber avoided exploiting out-of-scope services in 99.93% of ExploitGym Honeypot tests without its deployment safety prompt (up from 94.49% for GPT-5.6 Cyber), and 100% with it. That is a test of staying off services outside the assigned task.

The Firefox and Linux counts are research done with GPT-6 Cyber. OpenAI said that research produced evidence for 81 novel Firefox vulnerabilities that Mozilla classified as high severity, and showed that 30 known issues were more severe than previously reported. It said Mozilla has released fixes for many of them and began landing fixes after the reports. It did not say every finding is fixed. The same research developed working privilege-escalation exploits for 948 Linux vulnerabilities, after reproducing 1,724. OpenAI said that was more than five times the 166 it attributed to GPT-5.6 Sol on the same harness, and that 304 critical findings were disclosed to maintainers with candidate patches.

Access stays inside Daybreak Red. OpenAI said the model starts with approved customers on the API and Codex, with a wider Daybreak Red rollout planned. It also said GPT-5.6 Cyber is being opened to all Daybreak Red customers. The company named Cisco, Calif, and CodeRabbit as teams using it for vulnerability research and code review. It did not give a price. A separate OpenAI post set a Cyber at the Frontier livestream for November 12 at 10 a.m. Pacific, which is 2 p.m. AST.