OpenAI announced the launch of its cybersecurity model, 'GPT-5.6 Cyber,' which is specialized for security testing. The 'Daybreak' program, which provides cybersecurity models to approved individuals and organizations, operates in two tiers: Blue and Red.
Daybreak Blue utilizes the 'GPT-5.6 Sol' with safety measures in place to support approved defensive operations. OpenAI recommends that organizations performing general security tasks prioritize the use of Daybreak Blue.
Daybreak Red offers GPT-5.6 Cyber to organizations engaged in advanced vulnerability research, exploit development, and red team activities.
Both programs are available only to pre-approved individuals and organizations. Controls are in place during usage, including identity verification, account security, usage monitoring, restrictions on non-approved purposes, and compliance with relevant regulations.
According to OpenAI, GPT-5.6 Cyber achieved the highest success rate in 'ExploitGym,' a platform that evaluates the ability to implement known vulnerabilities into actual attack code. In assessments of advanced cybersecurity tasks, it completed 95% of all requests.
The model has also been used to identify vulnerabilities in actual software. While analyzing Chrome's JavaScript engine, 'V8,' OpenAI discovered two undisclosed vulnerabilities that could be exploited in conjunction.
After verifying the vulnerabilities, OpenAI communicated the findings to Google following a coordinated vulnerability disclosure process, and Google has since addressed the issues. The research process for actual vulnerabilities requires the ability to hypothesize and validate while analyzing large and unfamiliar codebases, as well as tracking interactions among various components to reproduce unexpected behaviors.
OpenAI is also expanding its Daybreak Cyber Partner Program in collaboration with major security and service companies, including Accenture, Cisco, Cloudflare, CrowdStrike, IBM, and Palo Alto Networks' 'Unit 42.' Daybreak Blue and Red will also be available to customers who have completed the registration process for Amazon Bedrock. Customers can utilize the model within their existing Amazon Web Services (AWS) security and governance environments.
Meanwhile, according to WiseApp and Retail, the most used AI chatbot app in South Korea last month was ChatGPT, which recorded 26.37 million monthly active users (MAU).
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.