What happened
On June 22, 2026, OpenAI expanded Daybreak with: (1) full GA of GPT-5.5-Cyber — achieving 85.6% on CyberGym (highest ever single-model score), 39.5% on ExploitGym, 69.8% on SEC-bench Pro, gated to vetted orgs via Trusted Access for Cyber (partners: Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, Oracle, Palo Alto Networks, Zscaler + 7 governments); (2) Patch the Planet, a coordinated OSS vuln-research program with Trail of Bits and HackerOne covering curl, Go, Python, Sigstore and 27 other projects; (3) Codex Security plugin (30M+ commits scanned across 30,000 codebases since March preview).
Why it matters
GPT-5.5-Cyber is the first AI model specifically trained and access-gated for defensive cyber operations at enterprise scale. Patch the Planet pairs AI-accelerated vuln discovery with mandatory human expert review before disclosure — a responsible AI-assisted security research model. The Codex Security scan scale (30M commits) signals what AI-native AppSec pipelines will look like.
Applicability
Security vendors in the Trusted Access for Cyber program gain immediate access; all enterprises should track Patch the Planet findings for OSS dependencies in their SBOMs.