Overview
- OpenAI unveiled the full GPT-5.5-Cyber under its Daybreak program and expanded Codex Security to integrate specialist models into defensive workflows.
- The company launched Patch the Planet and a Daybreak Cyber Partner Program to fund researchers, support open-source maintainers, and embed model outputs in vendor products.
- OpenAI reported leading scores on security tests, citing CyberGym 85.6%, ExploitGym 39.5%, and SEC-bench Pro 69.8% to show improved vulnerability-finding and exploit-generation ability.
- Developers disclosed a severe Codex bug that caused heavy TRACE logging to a local SQLite file at multi-megabyte-per-second rates, a pattern that could exhaust consumer SSD write lifespans, and OpenAI researcher Vaibhav Srivastav said a patched release is available with upgrade instructions.
- OpenAI says access will remain controlled to verified defenders while it coordinates deployment testing with government bodies and security vendors to balance faster remediation with operational and safety risks.