AI
OpenAI Launches ChatGPT Cyber and Expands Cybersecurity Model Program
OpenAI released ChatGPT Cyber for cybersecurity, available to users at the highest Daybreak Red tier. The move came after a startup technology executive used an OpenAI model to discover a critical Ubuntu-related vulnerability; OpenAI also said the same model scanned the Linux kernel and found 948 potential issues, and identified 81 new vulnerabilities in Firefox.
Max Justicz, co-founder and chief technology officer of Charge Robotics, told TechCrunch that he used GPT 5.6 Sol, released in July, to discover a critical vulnerability through the defender-oriented Daybreak Blue program. If exploited, attackers could potentially control servers running Ubuntu. He had the model review the same codebase where an error had occurred years earlier, and received results about two hours later. The Debian Project has released a patch, and there is currently no known evidence of exploitation.
Justicz said the vulnerability was highly convoluted and difficult for humans to discover. He said he spent roughly $5,000 on subscriptions and tokens while searching for multiple vulnerabilities, and expects serious vulnerabilities to be discovered and exploited in the coming months. He also noted that Daybreak Blue appears to have been weakened recently, which may constrain malicious users but also limits security researchers’ capabilities.
OpenAI announced that ChatGPT Cyber will be offered to Daybreak Red, the highest tier of the Daybreak program, which allows offensive security researchers to find and develop vulnerabilities. The company said the same model previously scanned and prepared patches for Linux kernel vulnerabilities, finding 948 potential issues, and also discovered 81 “new” vulnerabilities in Firefox. OpenAI also announced a $1000000000 program to support cybersecurity personnel participating in the program and protecting critical infrastructure such as power grids and water systems.