OpenAI has unveiled Daybreak, an initiative aimed at moving cyber defence beyond vulnerability discovery and into automated patching at scale. The programme combines an upgraded Codex Security plugin, a more capable and permissive GPT-5.5-Cyber model, a new partner programme, and an open-source remediation effort named Patch the Planet.
Codex Security has been in research preview since March and, by OpenAI's own account, has now swept through upwards of 30,000 codebases and 30 million commits. Of the vulnerabilities it has surfaced, human reviewers have signed off on more than 70,000 as genuinely fixed, while the system has self-certified another half a million or so without manual review. The updated plugin now offers full codebase scanning, threat modeling, and patch generation built directly into developer workflows.
The newly released GPT-5.5-Cyber, previously limited to a permissive-only preview, posts what OpenAI calls its highest CyberGym score yet at 85.6 percent, up from 81.8 percent for GPT-5.5, with similar gains reported on exploit-generation and long-horizon vulnerability discovery benchmarks. Access remains restricted to verified defenders under enhanced monitoring and review.

A Daybreak Cyber Partner Programme will give security vendors including Cisco, CrowdStrike, Palo Alto Networks and Wiz controlled access to the underlying models within their own products. Meanwhile Patch the Planet, run with Trail of Bits and HackerOne, has signed up more than 30 open-source projects, among them cURL, Go and Python, to receive expert-led remediation support.
OpenAI says it has struck Trusted Access for Cyber partnerships over the past month with Australia, Canada, France, Germany, Japan and South Korea, along with EU bodies including ENISA, alongside a separate and growing relationship with the UK government covering testing, evaluation and related work.


