OpenAI Launches Daybreak AI Security Program to Boost Vulnerability Detection

OpenAI launched Daybreak on May 11, 2026, combining Codex Security, GPT-5.5 and the specialized GPT-5.5-Cyber model to help enterprises build threat models, validate attack paths and generate patches. The move responds to Anthropic's Project Glasswing and underscores how the cybersecurity race has expanded into vulnerability validation, disclosure and patch deployment.

OpenAI expanded Daybreak on June 22, 2026, partnering with Trail of Bits to launch Patch the Planet. More than 30 open-source projects participated, and five-day sprints involving 19 projects identified hundreds of issues and merged dozens of patches. Results included eight information-disclosure proofs of concept and 24 local privilege-escalation exploits for the Linux kernel, as well as 34 confirmed FreeBSD vulnerabilities. The program did not disclose any funding amount.