OpenAI is launching GPT-6 Cyber, the latest version of its cybersecurity AI model, which it says is more effective at detecting vulnerabilities and less likely to make mistakes and hack other platforms in the process. OpenAI says that in testing, GPT-6 Cyber went out of bounds less than 1% of the time, compared to about 5% with GPT-5.6 Cyber.
The evaluation, called ExploitGym HoneyPot, was inspired by the Hugging Face incident, in which OpenAI’s AI agents breached the open source AI community to find answers to a problem they were charged with solving. “With Model 6 Cyber, we were able to make huge improvements to this assessment,” Fouad Matin, head of OpenAI’s cyber readiness and security team, said during a press briefing. “It’s more than 99 percent of the time, doing what you ask.”
But evaluation awareness, that is, when an AI system recognizes that it is being tested and changes its behavior in response, remains a concern. “Assessment awareness is a priority for us,” Matin said. “Often when we see assessment awareness, we see it in the model’s chain of thought. Matin added that OpenAI tries to prioritize production-style evaluations, in which the model interacts with real codebases with the real goal of reducing awareness risk.
GPT-6 Cyber is classified as Cybersecurity Critical under the OpenAI Readiness Framework, indicating that it “can identify and develop functional zero-day exploits of all severity levels in many real-world hardened critical systems without human intervention.” GPT-5.6 Cyber was previously classified as high in the framework.
GPT-6 Cyber outperforms GPT-6 Astra, OpenAI’s flagship critical-level model, on most cybersecurity tasks. In ExploitGym, an AI benchmark that measures whether an AI model can exploit vulnerabilities, GPT-6 Cyber scored 48.5%, about 16 points higher than GPT-5.6 Cyber’s 32.1% and a few points higher than GPT-6 Astra’s 42.4%. The model helped uncover 81 previously unknown vulnerabilities in Firefox and played a role in strengthening OpenAI’s own infrastructure for training AI models, OpenAI says.
Additionally, GPT-6 Cyber comes with stronger protections “to detect and disrupt misuse” while better respecting user instructions. OpenAI says it uses new controls to block “high-risk activities,” including attacks on critical infrastructure such as power plants and water treatment facilities.
GPT-6 Cyber is rolling out to select members of OpenAI’s Daybreak Red program, which is available to businesses and organizations using its models for “authorized defensive cybersecurity work.” GPT-6 Cyber is rolling out to all Daybreak Red customers in the coming weeks.
Gn tech
