OpenAI expands Daybreak program with GPT-5.6-Cyber for cybersecurity
OpenAI launched a two-tier Daybreak program providing access to advanced models for authorized security professionals. The new GPT-5.6-Cyber model is specifically trained for cybersecurity tasks and significantly increases completion rates for exploit development and vulnerability discovery requests. OpenAI has already used it successfully to find new vulnerabilities, including CVE-2026-15903 in Chrome.
Probably accurate. But if some companies learn how powerful models can be obtained, the models themselves will likely become targets of attacks aimed at acquiring them for their own use.
What are the key differences between Daybreak Blue and Daybreak Red?
Daybreak Blue provides access to general-purpose models (including GPT-5.6 Sol) without system-level safeguards, designed for standard defensive work like vulnerability discovery and malware analysis. Daybreak Red offers specially trained models including GPT-5.6-Cyber for advanced vulnerability research, exploit validation, and penetration testing with minimal restrictions.
Why does OpenAI create models with fewer safety restrictions?
Security specialists need access to advanced capabilities for defensive work that standard safety guardrails block. OpenAI aims to give legitimate defenders access to cutting-edge technology before attackers obtain it for offensive purposes. All models are provided only to authorized parties under oversight.
What results did OpenAI achieve with GPT-5.6-Cyber in finding real vulnerabilities?
OpenAI used GPT-5.6-Cyber to discover two previously unknown vulnerabilities in Chrome's V8 JavaScript engine that could be chained to escape the sandbox. Google assigned one as CVE-2026-15903 with high severity, and the fix has already been released.
- GPT-6 Astra: OpenAI's new AI model with advanced computer interaction — openai.com 75 % match
- OpenAI's Astra model reaches critical level of cybersecurity capabilities — openai.com 75 % match
- GLM-5.3 from Z.ai leads in vulnerability detection and will soon be available for download — z.ai 75 % match
- OpenAI26
- GPT-5.6-Cyber
- Daybreak
- GPT-5.6 Sol8
- V8
- Chrome8
- Google37
- CVE-2026-15903
- Jared Atkinson
- SpecterOps