SAN FRANCISCO — OpenAI on Monday expanded Daybreak, its cybersecurity initiative, giving participants broader access to AI capabilities for defensive work. The program now includes two tiers: Daybreak Blue and Daybreak Red.
Daybreak Blue offers access to OpenAI's advanced general-purpose models with adjusted safeguards for defensive security work. Daybreak Red provides purpose-trained cybersecurity models for security testing, vulnerability research and exploit validation.
OpenAI also launched GPT-5.6-Cyber, available to Daybreak Red users. The model builds on GPT-5.6 Sol, the company's most capable publicly available offering, and is designed to reduce refusals on specific cybersecurity tasks.
The expansion follows cybersecurity incidents disclosed by OpenAI, Anthropic and Meta in which AI models accessed systems that were off limits during security testing.
OpenAI first introduced Daybreak in May, positioning the program as a way for ecosystem partners to use its AI models to respond to a changing threat landscape.
The move comes after Anthropic launched its own cybersecurity coalition, Project Glasswing, which drew attention from Wall Street and the U.S. government.
OpenAI said it aims to put frontier intelligence into the hands of trusted defenders before attackers deploy offensive AI at scale. The company recommends Daybreak Blue as the starting point for most organizations seeking to use its AI for defensive purposes.
OpenAI last week paused some internal activities involving an upcoming model, Astra, after testing showed advances in agentic coding and cybersecurity. The company said it is assessing Astra's capabilities and implementing stronger safeguards and security controls, and committed to working with governments, safety institutes and civil society before deploying frontier capabilities.
