OpenAI launches GPT-5.5-Cyber for vulnerability patching

What happened

OpenAI launched the full version of GPT-5.5-Cyber, a model designed to find and patch software vulnerabilities, along with an updated Codex Security plugin that automates the full workflow from flaw discovery to patch generation. The company also established the Daybreak Cyber Partner Program with security firms including Cisco, CrowdStrike, Cloudflare, and Palo Alto Networks, as well as partnerships with governments including Australia, Canada, France, Germany, Japan, South Korea, and the UK.

Why it matters

According to OpenAI, the bottleneck in cybersecurity has shifted from identifying flaws to actually patching them. GPT-5.5-Cyber leads on key benchmarks—scoring 85.6% on CyberGym, 39.5% on ExploitGym, and 69.8% on SEC-bench Pro—outperforming Anthropic's Mythos 5 on the measures where both were tested. The updated Codex Security plugin has already scanned over 30 million commits and flagged over 500,000 findings, suggesting practical utility for developers at scale.

What to watch

GPT-5.5-Cyber access is deliberately restricted to verified defenders only and is tied to verification, monitoring, and guardrails. OpenAI recommends most users pair GPT-5.5 with Trusted Access for Cyber and Codex Security instead. An open-source patching initiative called Patch the Planet, launched with Trail of Bits, HackerOne, and Calif, has already signed on more than 30 open-source projects including cURL, Go, and Python.