The Strategic Shift to Proactive Defense

As threat actors increasingly adopt autonomous AI for cyberattacks, the window for defenders to respond is shrinking. OpenAI’s strategy is to provide "frontier intelligence" to trusted defenders before offensive capabilities reach mass scale. The expansion of the Daybreak program introduces two distinct tiers of access to balance safety with the high-performance requirements of security professionals.

Tiered Access and Model Specialization

The Daybreak program now offers two levels of access, each with different guardrail configurations:

  • Daybreak Blue: Removes standard system-level guardrails to facilitate incident detection, response, and general security assessments. It is recommended as the starting point for most defensive teams.
  • Daybreak Red: Provides access to GPT-5.6-Cyber, a model specifically trained to handle dual-use cybersecurity tasks. This model significantly reduces refusal rates for complex operations—completing 95% of requests involving exploit-chain development, authentication bypass, and privilege escalation, compared to 1.5% for the standard GPT-5.6 Sol model.

Performance in Real-World Security Workflows

GPT-5.6-Cyber is designed to excel in tasks requiring sustained reasoning across large, unfamiliar codebases. Its capabilities were demonstrated by the discovery of two high-severity vulnerabilities in the V8 JavaScript engine (CVE-2026-15903), which were subsequently reported and patched by Google.

Key performance metrics include:

  • ExploitGym: The model outperforms both GPT-5.6 Sol and GPT-5.5-Cyber in turning known vulnerabilities into working exploits within controlled environments.
  • Zero-Day Discovery: Specialized training allows the model to identify and calibrate the severity of novel vulnerabilities more effectively than the base model.
  • Trade-offs: While highly capable, the model can occasionally produce less detailed reports than the base model, and its performance on complex tasks like V8 exploitation can be token-intensive, requiring longer reasoning budgets (e.g., 600 turns vs. 300) to achieve optimal results.

Governance and Safety

Despite the removal of certain guardrails, OpenAI maintains strict control over Daybreak access through identity verification, legal attestations, and continuous monitoring. The models are assessed under OpenAI’s Preparedness Framework; while GPT-5.6-Cyber reaches the "High" threshold for cybersecurity capability, it remains below the "Critical" threshold, allowing for its release to authorized partners.