OpenAI Expands Daybreak Cybersecurity Push with GPT-5.5-Cyber, Codex Security and 'Patch the Planet'

OpenAI Goes All-In on Cyber Defence

OpenAI on Monday announced a sweeping expansion of its Daybreak cybersecurity initiative, unveiling a full version of its specialised GPT-5.5-Cyber model, an updated Codex Security plugin, a new partner programme for security vendors, and an open-source protection effort named "Patch the Planet." The announcement marks one of the company's most aggressive moves yet into defensive security tooling.

A Sharper, More Capable Model

The upgraded GPT-5.5-Cyber scored 85.6% on OpenAI's internal CyberGym benchmark, up from 81.8% for its predecessor GPT-5.5. On ExploitGym, the model jumped to 39.5% from 25.95%, and on SEC-bench Pro it reached 69.8% versus 63.1%. OpenAI says the model is built specifically for verified defenders whose work demands advanced cyber capabilities paired with verification, monitoring, and scoped controls, an acknowledgement of the dual-use risks that come with powerful offensive-security tooling.

Codex Security at Scale

The refreshed Codex Security plugin is designed to detect, validate, and remediate software vulnerabilities at scale. First previewed in March, Codex Security identifies flaws, proposes solutions, and fixes bugs, freeing developers to focus on higher-level engineering work rather than chasing individual defects.

Patch the Planet

Perhaps the most eye-catching piece is "Patch the Planet," launched in partnership with security research firm Trail of Bits. The initiative uses GPT-5.5-Cyber to hunt for critical bugs in open-source software. In its first week alone it surfaced hundreds of bugs, submitted 64 pull requests, and filed 51 issues across 19 projects. The programme also works alongside vulnerability management platforms such as HackerOne to provide free security consulting to open-source maintainers, many of whom are unpaid volunteers struggling to keep pace with a flood of AI-discovered flaws.

A Coalition of Security Heavyweights

The new Daybreak Cyber Partner Programme brings together a who's-who of the security industry, including Check Point, Tenable, TrendAI, Darktrace, Accenture, Cisco, CrowdStrike, IBM, and Palo Alto Networks. Participating partners can embed GPT-5.5 with Trusted Access for Cyber directly into their products and services, pushing the technology out to the defences relied upon by hundreds of thousands of organisations worldwide.

The Competitive Backdrop

Daybreak is widely seen as OpenAI's answer to Anthropic's Project Glasswing and its Mythos AI model, which debuted to limited preview earlier this year and has helped identify vulnerabilities across various industries. Crucially, unlike Mythos, Daybreak is publicly available, and companies can request security assessments directly from OpenAI. The company says it plans to expand the partner programme to additional organisations in the coming months, signalling that the race to put frontier AI at the heart of cyber defence is only just heating up.