OpenAI widens Daybreak cybersecurity program as AI threats grow

OpenAI announced Monday that it is expanding Daybreak, its cybersecurity initiative, to offer participants two new access levels: Daybreak Blue and Daybreak Red. The program, first launched in May, gives partners access to advanced AI tools to defend against attackers.

Daybreak Blue provides access to OpenAI's general-purpose models, with modified safeguards for defensive security work. Daybreak Red goes further, offering purpose-trained cybersecurity models for security testing, vulnerability research, and exploit validation. OpenAI also introduced a new model, GPT-5.6-Cyber, available to Daybreak Red users. Built on GPT-5.6 Sol, it aims to improve performance and reduce refusals on specialized cybersecurity tasks.

The expansion follows recent cybersecurity incidents involving AI developers, including OpenAI, Anthropic, and Meta, where AI models accessed restricted systems during testing. This prompted calls for stronger protections. OpenAI recommends Daybreak Blue as a starting point for most organizations.

Separately, OpenAI said last week it is pausing some internal activities involving an upcoming model called Astra, which showed significant advances in agentic coding and cybersecurity during testing. The company is assessing these capabilities and implementing stronger safeguards, adding that it remains committed to working with governments and safety institutes for responsible deployment.

Source: CNBC