OpenAI has unveiled its AI model, 'GPT-6 Cyber,' designed to assist in discovering and validating software vulnerabilities and developing security patches.
Notably, this model identified 81 previously unknown high-risk vulnerabilities in the Firefox web browser. Mozilla received the verification data and has addressed several of these vulnerabilities. In research on the Linux kernel, 1,724 vulnerabilities were reproduced, with functional attack codes developed for 948 of them.
The success rate of the 'Exploit Gym' in assessing the exploitability of vulnerabilities is 48.5%, an increase of 16.4 percentage points compared to previous models. In a separate evaluation, the rate of attacks on unauthorized services decreased from 5.51% to 0.07%.
GPT-6 Cyber is initially available to select approved Daybreak Red customers via API and codecs, with plans to expand access to all Daybreak Red customers in the future.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.