OpenAI has announced an update to its security-focused AI 'GPT-5.5-Cyber,' which surpasses Claude Mythos 5, and has also updated its security-focused Codex plugin 'Codex Security.'
OpenAI announced an update to its AI model for security researchers, ' GPT-5.5-Cyber, ' on June 22, 2026. The updated GPT-5.5-Cyber has achieved a score higher than Claude Mythos 5 in benchmark tests. An update to ' Codex Security ,' a plugin that adds security measures to Codex, was also announced.
Daybreak: Tools for securing every organization in the world | OpenAI
GPT-5.5-Cyber is an AI model provided to a select group of users who have passed OpenAI's review process, and is characterized by its high ability to discover program vulnerabilities. A preview version of GPT-5.5-Cyber has been available for some time, and this update enables 'detailed analysis across large codebases.' This allows for the development, testing, and preparation of patches for vulnerable code for human review.
The following are the results of testing the capabilities of 'GPT-5.5-Cyber (updated version)', 'Claude Mythos 5', 'GPT-5.5-Cyber (pre-update)', 'GPT-5.5', 'GPT-5.4', and 'Claude Opus 4.7' using the security capability benchmark test 'CyberGym'. The updated version of GPT-5.5-Cyber recorded a higher score than Claude Mythos 5.
To use GPT-5.5-Cyber, you need to pass OpenAI's review process. Individual users can apply for the review via the following link.
Trusted access for cybersecurity | ChatGPT
Additionally, an update to ' Codex Security ,' a plugin for Codex, was announced. Codex Security is an agent that assists in developing patches for vulnerabilities, and this update enables operations such as 'detailed code scanning,' 'displaying review results on a dashboard,' 'analyzing severity and impact,' 'generating reports including remediation guidance,' and 'building threat models.'
Furthermore, OpenAI announced the '
Related Posts:
