OpenAI is launching GPT-6 Cyber, a new cybersecurity AI model that the company says is more effective at identifying vulnerabilities while being less likely to hack other platforms. According to OpenAI, the model went out of bounds less than 1 percent of the time during testing, compared to approximately 5 percent with GPT-5.6 Cyber.
Fouad Matin, who leads OpenAI’s cyber preparedness and safety team, stated during a press briefing that the company made huge improvements on the ExploitGym HoneyPot evaluation. Matin said, “With the 6 Cyber model, we were able to make huge improvements on this evaluation,” and noted that the model is “over 99 percent of the time doing what you ask.”
The ExploitGym HoneyPot evaluation is modeled after the Hugging Face Incident, where OpenAI’s AI agents breached the open-source AI community to find answers to a specific problem. On this benchmark, GPT-6 Cyber scored 48.5 percent, which is about 16 points higher than the 32.1 percent scored by GPT-5.6 Cyber and a few points higher than the 42.4 percent scored by GPT-6 Astra.
OpenAI classifies GPT-6 Cyber as "Critical" on its Preparedness Framework. This classification means the model “can identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention.” This is an increase from GPT-5.6 Cyber, which was previously classified as High. GPT-6 Cyber also outperforms GPT-6 Astra, OpenAI’s flagship Critical-level model, on most cyber-related tasks.
Matin noted that evaluation awareness—where an AI recognizes it is being tested and changes its behavior—remains a concern. He stated, “Evaluation awareness is super top of mind for us,” and added that it often appears in the model’s chain of thought. To lessen this risk, Matin said OpenAI is trying to prioritize production-like evaluations where the model interacts with real codebases and real objectives.
The company says the model helped identify 81 previously unknown vulnerabilities in Firefox and assisted in hardening OpenAI’s own infrastructure for training AI models. OpenAI also stated that GPT-6 Cyber includes stronger safeguards to detect and disrupt misuse, including new checks to block high-risk activity such as attacks on water treatment facilities and power plants.
