OpenAI unveils GPT 6 Cyber model after security tests find software flaws

OpenAI announced GPT-6 Cyber on Thursday, a new artificial intelligence model designed for authorized security testing. BleepingComputer and The Hacker News both report that the model discovered 81 previously unknown high-severity Firefox vulnerabilities. BleepingComputer also reported that the model developed working exploits for 948 distinct Linux kernel flaws. SecurityWeek and Gulf Business both reported that the model has received a Critical cybersecurity rating. [1][3][4][5][6]

According to Help Net Security, GPT-6 Cyber is a version of GPT-6 Astra that was trained to identify software vulnerabilities, test exploit methods, and help developers create patches. Help Net Security reported that the model is designed for testing websites and applications, and OpenAI has used it to secure its own systems and AI model training infrastructure. BleepingComputer added that the model improves on earlier artificial intelligence versions by validating flaw impacts and assisting with fixes. [1][2]

GPT-6 is a family of large language models developed by OpenAI. GPT-6 Astra was released to the general public on September 4, 2026. The Hacker News reported that GPT-6 Cyber represents a major step forward in security research, combining stronger cybersecurity capabilities with improvements in model alignment. SecurityWeek reported that the model performs better at exploitation than GPT-5.6 Cyber while remaining within defined testing scopes. [5][6][10]

Technology Org reported that OpenAI is releasing the GPT-6 Cyber model to vetted security defenders. The Hacker News reported that the launch brings together defensive capabilities with enhanced model safeguards. [5][9]