OpenAI has significantly expanded its Daybreak program by introducing two new access tiers—Daybreak Blue and Daybreak Red—alongside the release of GPT-5.6-Cyber, a specialized AI model designed for exploit validation, vulnerability research, and red teaming.
Daybreak Blue serves as the entry point for most security defenders, providing access to GPT-5.6 Sol. This model is tailored for tasks such as vulnerability discovery, secure code review, malware analysis, incident response, and patch validation, with safeguards in place to ensure responsible use.
Daybreak Red offers a more advanced level of access, granting vetted security professionals the ability to utilize GPT-5.6-Cyber. This model is specifically trained to enhance performance in zero-day discovery and exploit-chain development, while reducing refusals on legitimate but high-risk security prompts, such as penetration testing in production environments.
In internal benchmarks, GPT-5.6-Cyber demonstrated a 95% completion rate for complex security tasks, a significant improvement over the 57.3% achieved by its predecessor, GPT-5.5-Cyber. This advancement addresses previous concerns from security researchers regarding excessive model refusals during legitimate security operations.
OpenAI utilized GPT-5.6-Cyber to investigate Chrome’s V8 JavaScript engine, uncovering two previously unknown vulnerabilities. These findings were disclosed to Google and patched as CVE-2026-15903, a high-severity flaw that could have allowed attackers to execute arbitrary code within Chrome’s sandbox.
To mitigate potential misuse risks associated with reduced safeguards, OpenAI is implementing several security measures. Starting September 1, 2026, hardware security keys will be mandatory for all individual Daybreak accounts. Additionally, Codex users are being encouraged to adopt auto-review mode over full-access mode, and enhanced monitoring protocols are being rolled out.
Access to both Daybreak Blue and Red is restricted to approved individuals and organizations conducting authorized security work. This access is contingent upon identity verification, monitoring, and legal attestations. OpenAI recommends that users sandbox workflows, tightly scope permissions, and maintain human oversight for higher-risk tasks.
Security firms, including SpecterOps, have reported significant workflow acceleration using GPT-5.6-Cyber. SpecterOps CTO Jared Atkinson noted that the model resolved specialist vulnerability research tasks in under a day, a process that previously took weeks.
Organizations interested in Daybreak Red can apply through OpenAI’s partner program.
OpenAI’s expansion of the Daybreak program with GPT-5.6-Cyber marks a significant advancement in AI-driven cybersecurity. By providing vetted security professionals with powerful tools for exploit validation and vulnerability research, OpenAI is addressing the growing asymmetry between attackers and defenders in the digital landscape. However, the implementation of stringent access controls and monitoring underscores the importance of balancing innovation with security to prevent potential misuse.
