OpenAI has announced the expansion of its Daybreak program, introducing GPT-5.6-Cyber to provide frontier cyber models for defenders. This initiative aims to equip trusted defenders with advanced AI capabilities before attackers widely deploy offensive AI.
The Daybreak program now features two access tiers, Daybreak Blue and Daybreak Red, each tailored to different defensive needs. GPT-5.6-Cyber, available through Daybreak Red, is built upon GPT-5.6 Sol and is specifically trained to improve performance on cybersecurity tasks such as finding zero-day vulnerabilities and developing exploit chains. It also reduces refusals for certain dual-use cyber tasks.
Key Points
- OpenAI expanded its Daybreak program with two access tiers for approved defenders.
- GPT-5.6-Cyber is introduced, built on GPT-5.6 Sol, and available through Daybreak Red access.
- GPT-5.6-Cyber is trained to improve capabilities in specialized cybersecurity tasks, including finding zero-day vulnerabilities and developing exploit chains.
- The model aims to reduce refusals for certain higher-risk, dual-use cyber tasks.
- Daybreak Blue access removes system-level safeguards from GPT-5.6 Sol, which can block legitimate defensive work.
- GPT-5.6-Cyber completes 95.0% of requests in an internal Advanced Cybersecurity Completion Rate evaluation, compared to 1.5% for GPT-5.6 Sol and 2.0% for GPT-5.6 Sol with Daybreak Blue access.
- GPT-5.6-Cyber also outperforms GPT-5.5-Cyber, which completed 57.3% of requests in the same evaluation.
Context
According to OpenAI Security, the cybersecurity landscape is evolving rapidly, with threat actors increasingly using AI for cyberattacks. The Daybreak program is designed to provide trusted defenders with advanced intelligence to prepare for these changes. GPT-5.6 Sol delivers state-of-the-art performance on cybersecurity tasks, but its production safeguards can impede legitimate defensive work. Daybreak Blue access addresses this by removing these guardrails for GPT-5.6 Sol in real-world security tasks, including incident detection, investigations, and vulnerability management. For highly dual-use prompts where GPT-5.6 Sol might still refuse, GPT-5.6-Cyber through Daybreak Red access is trained to further reduce refusals and improve performance.
Why It Matters
The introduction of GPT-5.6-Cyber and the tiered Daybreak program offers cybersecurity professionals enhanced tools to counter evolving AI-driven threats. By reducing model refusals and improving performance on specialized tasks, these models can accelerate defensive workflows, allowing researchers to focus more on validating findings and developing protective measures.
What To Do
- Review the capabilities of GPT-5.6-Cyber for tasks like exploit chain development and zero-day vulnerability discovery.
- Compare the refusal rates of GPT-5.6-Cyber with GPT-5.6 Sol and GPT-5.5-Cyber for advanced cybersecurity scenarios.
- Note the differences in access and guardrail removal between Daybreak Blue and Daybreak Red tiers.
- Consider how reduced refusals could impact vulnerability research and incident response workflows.