Particle.news
Download on the App Store

OpenAI Pushes from Finding Bugs to Fixing Them with GPT‑5.5‑Cyber and Patch the Planet

OpenAI says the initiative will close a growing remediation backlog by using a restricted GPT‑5.5‑Cyber plus automated Codex Security with staffed Patch the Planet support.

Overview

  • OpenAI expanded its Daybreak program on Monday by releasing GPT‑5.5‑Cyber to vetted defenders, updating the Codex Security plugin to generate and test patches, and launching a Daybreak Cyber Partner Program for major security vendors.
  • Patch the Planet, created with Trail of Bits and developed with HackerOne and CALIF, has already enrolled more than 30 open‑source projects and ran a five‑day sprint where 25 Trail of Bits engineers used Codex/GPT‑5.5‑Cyber to surface hundreds of issues and merge dozens of fixes.
  • OpenAI reports Codex Security has scanned over 30 million commits across about 30,000 codebases with human reviewers marking more than 70,000 findings fixed and the system auto-resolving over 500,000 findings.
  • GPT‑5.5‑Cyber shows measurable gains on security benchmarks (85.6% on CyberGym) and can analyze large codebases, trace attack paths, validate vulnerabilities in controlled settings, and generate codebase-specific patches for human review.
  • Security agencies have warned that frontier models speed both offense and defense, so OpenAI is keeping access gated and pairing automated tooling with human engineers to reduce false positives, limit risky misuse, and help maintainers adopt durable testing and patching workflows.