OpenAI has announced GPT-5.6-Cyber, a new cybersecurity-focused model built on GPT-5.6 Sol and trained to handle restricted dual-use prompts that standard models refuse. The model is available only through Daybreak Red, a new tier in OpenAI's gated cybersecurity program, alongside Daybreak Blue which opens GPT-5.6 Sol to defenders with system-level cyber guardrails removed. The launch addresses a persistent tension: frontier models' safeguards block legitimate defensive work like vulnerability discovery and exploit validation.
In OpenAI's Advanced Cybersecurity Completion Rate benchmark, GPT-5.6-Cyber responds to 95% of advanced security prompts involving exploit chains, authentication bypasses, and privilege escalation—compared to just 1.5% for standard Sol and 2% for Sol through Daybreak Blue. The refusal-reduction comes from specialized training on zero-day discovery and exploit-chain development. OpenAI tested the model on V8, Chrome's JavaScript engine, and reports it found two previously unknown vulnerabilities that could be chained to corrupt memory and escape the V8 heap sandbox, both disclosed to Google and assigned CVE-2026-15903.
Access to both Daybreak tiers requires identity verification, approved-use restrictions, monitoring, legal attestations, and hardware security keys (mandatory from September 1, 2026). Daybreak Red targets vulnerability researchers and red teamers; Blue is OpenAI's recommended entry point for broader defensive work. Partners including Accenture, IBM, CrowdStrike, Cisco, and Palo Alto Networks can embed the models into security products and managed services. For architects evaluating defense tooling, Daybreak's tiered structure signals how OpenAI is balancing capability access with governance—a model other labs are likely to follow.