OpenAI launches cybersecurity model GPT-6 Cyber: 48.5% on ExploitGym, far fewer out-of-scope attacks, first for select Daybreak Red customers
OpenAI released GPT-6 Cyber on 5 November, a cybersecurity variant of GPT-6 Astra built to find vulnerabilities, validate exploitability and help write patches while staying within authorized scope. According to The Verge and SecurityWeek, it scores 48.5% on ExploitGym and attacked off-limits decoy services in 0.07% of tests, down from about 5.5% for GPT-5.6 Cyber; for now it is available only through the API and Codex to a small group of approved Daybreak Red customers.

According to The Verge, GPT-6 Cyber scores 48.5% on ExploitGym, which tests whether a model can exploit a specified vulnerability, about 16 points above GPT-5.6 Cyber’s 32.1% and a few points above GPT-6 Astra’s 42.4%; OpenAI says it helped find 81 previously unknown vulnerabilities in Firefox. SecurityWeek describes a dedicated scope test in which the task sat beside services that were deliberately easy to hack but off-limits: without deployment safety prompts GPT-6 Cyber left them alone 99.93% of the time (94.49% for GPT-5.6 Cyber), and with the prompts it never touched them. The report adds that the model reproduced 1,724 Linux kernel vulnerabilities and built working privilege-escalation exploits for 948 of them, and that OpenAI has reported 304 critical issues to maintainers.
The Verge says GPT-6 Cyber is rated Critical in cybersecurity under OpenAI’s Preparedness Framework (GPT-5.6 Cyber was High) and ships with new checks that block high-risk activity against critical infrastructure such as power plants and water treatment facilities. OpenAI’s system card rates its biological and chemical capabilities High but not Critical, in line with GPT-6 Astra; API customers using Zero Data Retention generally need Private Safety Processing so encrypted safety reviews can run, and the industrial-control safeguards were evaluated with Dragos. A system-card chart shows completion rates of 100%, 96% and 97% on authorized vulnerability patching, exploit development and red-teaming tasks. The Verge says OpenAI plans to extend access to all Daybreak Red customers in the coming weeks.
Sources
- GPT-6 Cyber System Card — OpenAI
- OpenAI says its new cybersecurity model is better at not hacking other companies — The Verge
- OpenAI Says New GPT-6 Cyber Model Hunts Flaws Without Crossing the Line — SecurityWeek