OpenAI Expands Daybreak: GPT-5.5-Cyber (85.6% CyberGym) + "Patch the Planet" Remediation
Jun 22, 2026: OpenAI releases cyber-tuned GPT-5.5-Cyber (85.6% vs 81.8% CyberGym) to vetted defenders, updates Codex Security (30M commits / 30k codebases scanned), and launches Patch the Planet with Trail of Bits + HackerOne.
TL;DR — Jun 22, 2026: OpenAI expands Daybreak — full GPT-5.5-Cyber (85.6% vs 81.8% CyberGym; vetted access), updated Codex Security (30M commits / 30k+ codebases scanned), Cyber Partner Program, and Patch the Planet (Trail of Bits + HackerOne, human-in-loop). Thesis: shift AI security value from find → fix.
A defense-side play on remediation throughput.
Model
| Metric | GPT-5.5-Cyber | GPT-5.5 (standard) |
|---|---|---|
| CyberGym benchmark (reproduce known vulns) | 85.6% | 81.8% |
| Access | Vetted "trusted defenders" only | General |
| Purpose | Find, validate & patch vulnerabilities | General-purpose |
Facts
- Patch the Planet: founded w/ Trail of Bits, w/ HackerOne; human review before maintainer contact; 5-day sprint across 19 projects.
- Finds credited: 23-yr OpenBSD kernel use-after-free; 5 V8 bugs; WebKit/Safari issues.
- Partners: Cisco, CrowdStrike, IBM, Okta, Palo Alto, Proofpoint + govts (AU/CA/FR/DE/JP/KR). Source: SiliconANGLE, Infosecurity.
- Ryan Kalember, CSO, Proofpoint: integrating GPT-5.5 "can help security teams improve threat investigation, decision-making, and efficiency."
FAQ
Model edge?
GPT-5.5-Cyber 85.6% CyberGym vs 81.8% standard; vetted access.
Scale scanned?
30M commits / 30k+ codebases since Mar 2026.
Patch the Planet?
Open-source fixing w/ Trail of Bits + HackerOne, human-reviewed.
Real finds?
23-yr OpenBSD flaw; 5 Chrome V8 bugs.
Sources: SiliconANGLE, Infosecurity Magazine, IT Pro.
Image: mikemacmarketing, CC BY 2.0, via Wikimedia Commons.
← Back to all posts