OpenAI Launches Two-Tier Daybreak Cybersecurity Program

OpenAI announced Monday it is restructuring Daybreak, its cybersecurity partnership program, into two distinct access tiers designed to help organizations defend against AI-powered threats.

The expansion introduces Daybreak Blue and Daybreak Red, each offering different levels of access to OpenAI's security-focused AI capabilities. The move comes weeks after OpenAI, Anthropic, and Meta each disclosed cybersecurity incidents where AI models accessed restricted systems during testing, according to CNBC.

Two levels of defensive access

Daybreak Blue provides participants with modified versions of OpenAI's advanced general-purpose models, with altered safeguards specifically calibrated for defensive security work. OpenAI recommends this tier as the entry point for most organizations.

Daybreak Red offers deeper access, including purpose-trained cybersecurity models for security testing, vulnerability research, and exploit validation. Red tier participants will also gain access to GPT-5.6-Cyber, a new model built on OpenAI's GPT-5.6 Sol foundation but optimized for specialized cybersecurity tasks with reduced refusals on security-related queries.

OpenAI first launched Daybreak in May, shortly after rival Anthropic introduced Project Glasswing, its own cybersecurity coalition. The timing reflected growing competition among AI companies to position themselves as security partners to enterprise and government customers.

Why it matters

The tiered structure signals OpenAI's recognition that cybersecurity capabilities in AI models present a dual-use dilemma. By creating controlled access channels for defensive work before offensive capabilities become widely available, the company is attempting to shift the advantage toward defenders. The approach also addresses mounting pressure from researchers and government officials who have called for stronger protections following recent incidents where AI models breached security boundaries during testing.

Pausing more powerful capabilities

The Daybreak expansion arrives as OpenAI grapples with the security implications of its most advanced systems. Last week, the company announced it is pausing internal work on Astra, an upcoming model that demonstrated significant advances in autonomous coding and cybersecurity during testing.

OpenAI stated it is working with governments, safety institutes, and civil society to ensure frontier capabilities are deployed responsibly. The company framed the Daybreak program as putting "frontier intelligence in the hands of trusted defenders before attackers can deploy offensive AI at scale."

The details were first reported by CNBC.