Daybreak Full GPT-5.5-Cyber: Politically Safest Place to Ship Frontier Uplift

Cyber capability is becoming the politically safest place to ship frontier uplift: governments care about critical-infrastructure defense, and labs can argue dual-use tools belong with verified responders. Daybreak’s full cyber model plus open-source patching scale is OpenAI’s bid to own that narrative—before rivals set the default stack.

OpenAI pushed its Daybreak cyber-defense stack forward with two primary posts: the full GPT‑5.5‑Cyber release for trusted defenders, and Patch the Planet, a program to accelerate vulnerability discovery and patching in open source. GPT‑5.5‑Cyber is reported at 85.6% on CyberGym versus 81.8% for GPT‑5.5. Trail of Bits engineers working full-time with Codex and the cyber model across 19 open-source projects had already identified hundreds of security issues and merged dozens of patches.

GPT-5.5-Cyber full release

From OpenAI’s Daybreak update:

• Audience: Continued limited release to trusted defenders (not a free public offensive toolkit).
• Positioning: More permissive for authorized cybersecurity work and more capable at finding and helping patch vulnerabilities, while retaining GPT‑5.5 general intelligence for long tasks.
• Benchmark claim: 85.6% CyberGym vs 81.8% for standard GPT‑5.5 (vendor eval).
• Lineage: Follows an earlier permissive-only preview; part of OpenAI’s broader Trusted Access for Cyber posture.

Patch the Planet

From OpenAI’s Patch the Planet post (Daybreak initiative):

• Goal: Support open-source maintainers in finding, validating, and fixing vulnerabilities with AI assistance.
• Partner example: Trail of Bits dedicated security engineers full-time with Codex and GPT‑5.5‑Cyber across 19 projects—hundreds of issues identified, dozens of patches merged, more under coordinated disclosure.
• Early field notes (OpenAI-cited): fuzzing-lab construction in under a day that would “ordinarily take at least several weeks”; Linux kernel work including automated PoCs for pointer leaks and local privilege escalations among a larger issue set.

Together with related Codex security tooling, the package is OpenAI’s public answer to industry pressure—heightened after Anthropic’s Project Glasswing and the Fable/Mythos export fight—to put frontier cyber capability in defenders’ hands under access controls.

Limits

• CyberGym 85.6% is vendor-reported—not an independent public cyber Elo board.
• Trusted-defender limited release: not a free public offensive toolkit.
• Patch counts and “under a day” fuzzing-lab notes are OpenAI/Trail of Bits field claims.

Sources

• OpenAI: “Daybreak: Tools for securing every organization in the world” / GPT‑5.5‑Cyber full update (openai.com/index/daybreak-securing-the-world/, June 22, 2026).
• OpenAI: “Patch the Planet: a Daybreak initiative to support open source maintainers” (openai.com/index/patch-the-planet/, June 22, 2026).