OpenAI has expanded its cyber security program "Daybreak" into two tracks, "Daybreak Blue" and "Daybreak Red". It will provide Red users with a new model, "GPT-5.6-Cyber."

CNBC reported on Aug. 10 that OpenAI said the move would provide needed capabilities to trusted defenders before attackers use AI at scale.

Daybreak is a closed cyber security program OpenAI released in May. OpenAI said it introduced Daybreak to help ecosystem partners respond to a rapidly changing threat environment with its advanced AI models.

The program expansion comes after cyber security incidents in recent weeks triggered by AI models.

OpenAI, Anthropic and Meta models entered systems that they should not have accessed and that are not supposed to be entered, during security testing. Industry researchers and government officials are now calling for stronger safeguards.

OpenAI said Daybreak Blue supports access to OpenAI's advanced general-purpose models, with guardrails adjusted for defensive security tasks. OpenAI is recommending that most organisations start with Daybreak Blue.

Daybreak Red provides access to cyber security models trained for security testing, vulnerability research and exploit verification. Users in this tier can also use "GPT-5.6-Cyber."

"GPT-5.6-Cyber" is based on OpenAI's current top publicly available model, "GPT-5.6 Sol." OpenAI said it was designed to boost capabilities for specific specialised security tasks and reduce refusals on some tasks.