OpenAI expands Daybreak security tools

OpenAI expanded Daybreak, its cybersecurity line, on June 22 with three pieces a working team can actually feel. Codex Security, in research preview since March, has now scanned more than 30 million commits across 30,000-plus codebases, with human reviewers confirming over 70,000 findings as fixed. GPT-5.5-Cyber, a more capable security model, went out in a limited release to vetted defenders; OpenAI reports it scores 85.6% on the CyberGym benchmark against 81.8% for GPT-5.5. And Patch the Planet, run with Trail of Bits and HackerOne, puts AI and human engineers on widely used open-source projects: across 19 projects in its first week, the partners say they turned up hundreds of bugs, opened 64 pull requests, and filed 51 issues.

Why this matters

Most small teams ship on top of open-source they will never audit, so an effort that moves real fixes into those projects lowers a risk you carry without paying for it. The flip side is the model. GPT-5.5-Cyber is gated to verified defenders precisely because stronger cyber skill cuts both ways, so treat any “AI found a vulnerability” claim the way you would any AI output worth checking. If you run coding agents, this lands right next to recent warnings about fake bug reports hijacking agents and a Copilot flaw that leaked data.