OpenAI Launches GPT-6 Cyber with Enhanced Safety Safeguards and Improved Vulnerability Detection
OpenAI has introduced GPT-6 Cyber, its latest cybersecurity AI model designed to identify vulnerabilities while minimizing unauthorized hacking attempts. According to OpenAI, the new model demonstrated significantly improved safety, going 'out of bounds' less than 1% of the time during testing, a substantial decrease from the 5% rate seen in the previous GPT-5.6 Cyber version. The model has been classified as 'Critical' under OpenAI’s Preparedness Framework due to its ability to identify zero-day exploits in hardened systems without human intervention. In benchmark testing via ExploitGym, GPT-6 Cyber achieved a score of 48.5%, outperforming both GPT-5.6 Cyber and the flagship GPT-6 Astra model. The technology is currently being rolled out to select members of OpenAI’s Daybreak Red program for authorized defensive cybersecurity work.
Topics: AI safety cybersecurity OpenAI product launch