OpenAI introduced GPT-6 Cyber, a new security research model. The model scored 48.5% on the ExploitGym vulnerability benchmark. It also helped researchers develop 948 Linux privilege escalation exploits and find 81 high-severity Firefox bugs.
Benchmark gains
GPT-5.6 Cyber scored 32.1% on the same benchmark. In a separate alignment evaluation, GPT-6 Cyber avoided out-of-scope services in 99.93% of tests without deployment prompts. That rate rose to 100% with those prompts.
Adoption and use
Security teams at Cisco, Calif, and CodeRabbit are using the model for vulnerability research and code review. OpenAI also used GPT-6 Cyber to find and fix weaknesses in its own infrastructure. The company plans to expand its Daybreak program to more organizations and open-source maintainers.
Daybreak offers
Daybreak Blue helps verified defenders validate vulnerabilities and fixes. Daybreak Red supports deeper exploit development and authorized security testing. A new Daybreak Red group will serve approved critical infrastructure defenders.
OpenAI now classifies GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework. The company is expanding partnerships with security providers, energy companies, and critical infrastructure operators. It will continue refining safeguards using partner feedback and user reports.