OpenAI expands Daybreak cybersecurity program with GPT‑5.5‑Cyber, Patch the Planet and new enterprise partners

OpenAI announced the full release of GPT‑5.5‑Cyber, a defensive‑focused AI model that achieved an 85.6% score on the CyberGym benchmark, outperforming Anthropic’s Mythos 5. The model can analyse large codebases, validate vulnerabilities and generate patch proposals, while human security engineers retain final approval.

Alongside the model, OpenAI unveiled an updated Codex Security plugin that turns the Codex coding assistant into an application‑security platform, automatically scanning code, building threat models and suggesting fixes. A new initiative called Patch the Planet, created with Trail of Bits, HackerOne and Calif, will provide security engineers to work directly with maintainers of more than 30 high‑impact open‑source projects such as cURL, Python, Go, Sigstore and pyca/cryptography. In its first week the program reported hundreds of issues, dozens of pull‑requests and dozens of closed problems.

OpenAI also broadened its Daybreak Cyber Partner Program, adding major vendors and integrators—including IBM, Accenture, Cisco, CrowdStrike, Palo Alto Networks, Check Point and others—to embed the GPT‑5.5‑Cyber capabilities into enterprise security workflows. IBM’s participation builds on its $5 billion Project Lightwell effort to secure open‑source supply chains. Access to the GPT‑5.5‑Cyber model remains limited to vetted defenders through OpenAI’s Trusted Access for Cyber framework.