OpenAI launches Daybreak, a cybersecurity platform combining GPT-5.5 models and Codex Security to help defenders identify vulnerabilities and patch code faster than attackers can exploit them.

OpenAI introduced Daybreak, a cybersecurity initiative built around GPT-5.5 models and Codex Security, partnering with companies including Cloudflare, CrowdStrike, Cisco, and Palo Alto Networks. The system helps security teams identify vulnerabilities, validate fixes, generate patches, and monitor risky code.

Daybreak splits access into tiers: regular GPT-5.5 with standard safeguards, 'Trusted Access for Cyber' for verified defensive workflows like malware analysis and vulnerability triage, and GPT-5.5-Cyber for authorized red teaming and penetration testing. This approach differs from Anthropic's Mythos, which restricts access to a small group of companies under Project Glasswing.

The launch addresses a gap where AI-assisted vulnerability discovery is faster than most companies can patch: HackerOne temporarily paused parts of its bug bounty program earlier this year because maintainers were flooded with AI-assisted reports, and one security researcher argued the traditional 90-day disclosure window is essentially obsolete given how quickly multiple models can find the same vulnerability.