OpenAI Updates GPT-5.5-Cyber, Surpassing Claude Mythos 5 in Security Benchmarks
The update positions GPT-5.5-Cyber ahead of a key competitor in the security AI space, while the expanded plugin and partner program signal OpenAI's push to embed its security tools into enterprise workflows.
Key Facts
- OpenAI updated GPT-5.5-Cyber on June 22, 2026, a model restricted to users who pass the company's screening.
- The updated GPT-5.5-Cyber scored higher than Anthropic's Claude Mythos 5 on the CyberGym benchmark, a security capability test.
- OpenAI launched the Daybreak Cyber Partner Program with initial partners IBM, CrowdStrike, Accenture, and Akamai.
- OpenAI announced Patch the Planet, an initiative to collaborate with open-source project maintainers.
Reporting from 2 sources: ASCII.jp, GIGAZINE.
OpenAI announced an update to GPT-5.5-Cyber, its AI model for security researchers, on June 22, 2026. The updated model scored higher than Anthropic's Claude Mythos 5 on the CyberGym benchmark. OpenAI also updated the Codex Security plugin and launched the Daybreak Cyber Partner Program with partners including IBM and CrowdStrike.
OpenAI released an update to GPT-5.5-Cyber on June 22, 2026, a model restricted to users who pass the company's screening. The updated version scored higher than Anthropic's Claude Mythos 5 on the CyberGym benchmark, a security capability test. The model can now perform detailed analysis across large codebases, including patch development, testing, and preparation for human review.
OpenAI also updated Codex Security, a plugin that assists in vulnerability patching, adding features such as detailed code scanning, dashboard display of review results, severity and impact analysis, report generation, and threat model construction. The company announced the Daybreak Cyber Partner Program, a framework for integrating its security products into third-party services, with IBM, CrowdStrike, Accenture, and Akamai as initial participants. An initiative called Patch the Planet, aimed at collaborating with open-source project maintainers, was also announced.
Synthesized by Yomimono from the 2 cited sources below, including Japanese-language reporting where cited, then editorially reviewed before publishing.