It primarily targets organizations performing specialized security tasks such as advanced vulnerability research and red team activities.

Simultaneously, the company announced that it will operate the 'Daybreak' cybersecurity model delivery program in two distinct tracks: Blue and Red.

Daybreak Blue is based on GPT-5.6 Sol and applies safeguards tailored to defensive operations, while Daybreak Red provides GPT-5.6 Cyber.

Both tiers are provided exclusively to pre-approved users who undergo controls including identity verification, account security, monitoring, and legal compliance verification.

According to the company, GPT-5.6 Cyber achieved the highest success rate among models tested by OpenAI in the 'ExploitGym' evaluation, which measures the ability to implement known vulnerabilities into actual attack code.

The completion rate for advanced cybersecurity tasks reached 95% of all requests.

Real-world application examples were also disclosed.

OpenAI identified two previously undisclosed vulnerabilities that could be chained together while analyzing Chrome's JavaScript engine V8 using GPT-5.6 Cyber.

The findings were reported to Google following verification and coordinated disclosure procedures, and Google has patched the vulnerabilities.

OpenAI is expanding the Daybreak Cyber Partner Program with six security partners: Accenture, Cisco, Cloudflare, CrowdStrike, IBM, and Palo Alto Networks Unit 42.

Partner companies can integrate the model into their products, managed security services, and customer projects.

Amazon Bedrock-qualified customers who have completed Daybreak registration can utilize both model tiers within their existing AWS security and governance environment.

Program participation applications are available at openai.com/daybreak/partners.