ChatGPT
November 6, 2026
OpenAI Introduces GPT-6 Cyber
Following the discovery of 948 vulnerabilities in Linux, OpenAI has introduced GPT-6 Cyber, a specialized model designed for cybersecurity research.The model can analyze code, identify previously unknown vulnerabilities, assess their exploitability, and develop working exploits. OpenAI previously used the model to discover 81 high-severity, previously unknown vulnerabilities in Firefox.
Following the discovery of 948 vulnerabilities in Linux, OpenAI has introduced GPT-6 Cyber, a specialized model designed for cybersecurity research.The model can analyze code, identify previously unknown vulnerabilities, assess their exploitability, and develop working exploits. OpenAI previously used the model to discover 81 high-severity, previously unknown vulnerabilities in Firefox.
On ExploitGym, a benchmark that evaluates a model's ability to exploit a specified vulnerability in a remote system, GPT-6 Cyber achieved a success rate of 48.5%, compared with 32.1% for GPT-5.6 Cyber and 42.4% for GPT-6 Astra.In a separate black-box web vulnerability discovery evaluation, where models must identify security flaws without access to source code, GPT-6 Cyber detected 40.3% of vulnerabilities, outperforming GPT-6 Astra at 32%.The new model also demonstrated improved adherence to authorized testing boundaries. OpenAI evaluated how reliably it stayed within the permitted scope of security assessments without attempting to attack unrelated services.Without additional instructions, GPT-6 Cyber remained within its authorized scope in 99.93% of cases, compared with 94.49% for GPT-5.6 Cyber. With additional safety instructions, the new model did not cross the established boundaries in any of the tests.
Benchmark Results
On ExploitGym, a benchmark that evaluates a model's ability to exploit a specified vulnerability in a remote system, GPT-6 Cyber achieved a success rate of 48.5%, compared with 32.1% for GPT-5.6 Cyber and 42.4% for GPT-6 Astra.In a separate black-box web vulnerability discovery evaluation, where models must identify security flaws without access to source code, GPT-6 Cyber detected 40.3% of vulnerabilities, outperforming GPT-6 Astra at 32%.The new model also demonstrated improved adherence to authorized testing boundaries. OpenAI evaluated how reliably it stayed within the permitted scope of security assessments without attempting to attack unrelated services.Without additional instructions, GPT-6 Cyber remained within its authorized scope in 99.93% of cases, compared with 94.49% for GPT-5.6 Cyber. With additional safety instructions, the new model did not cross the established boundaries in any of the tests.