Expanding Daybreak as the Cyber Defense Window Narrows
OpenAI is widening Daybreak access and adding GPT-5.6-Cyber for approved defenders doing higher-risk security work.
The post says Daybreak now has Blue access for frontier general models and Red access for purpose-trained cyber models. GPT-5.6-Cyber is described as reducing refusals on advanced dual-use tasks, with OpenAI saying it completed 95.0% of requests in an internal advanced cybersecurity evaluation. OpenAI also says the model helped find real vulnerabilities, including a high-severity V8 issue reported to Google as CVE-2026-15903. Access is limited to approved users and organizations, with identity checks, monitoring, legal attestations, hardware security keys for individual accounts, and stronger review settings encouraged.
OpenAI's note
The post says Daybreak now has Blue access for frontier general models and Red access for purpose-trained cyber models. GPT-5.6-Cyber is described as reducing refusals on advanced dual-use tasks, with OpenAI saying it completed 95.0% of requests in an internal advanced cybersecurity evaluation. OpenAI also says the model helped find real vulnerabilities, including a high-severity V8 issue reported to Google as CVE-2026-15903. Access is limited to approved users and organizations, with identity checks, monitoring, legal attestations, hardware security keys for individual accounts, and stronger review settings encouraged.
OpenAI's note
score 7