OpenAI unveils GPT-5.6-Cyber for advanced cybersecurity with limited access
OpenAI introduced GPT-5.6-Cyber, a model fine-tuned for high-level vulnerability research and exploit development, available only to vetted enterprise defenders through its new Daybreak Red program.
OpenAI announced the launch of GPT-5.6-Cyber, a specialized version of its latest GPT-5.6 Sol model that has been fine-tuned for advanced cybersecurity tasks such as zero-day discovery, exploit-chain creation and privilege escalation. In internal testing, the model achieved a 95 % completion rate on OpenAI’s Advanced Cybersecurity Completion benchmark, compared with 57.3 % for the preceding GPT-5.5-Cyber and 1.5 % for the standard GPT-5.6 Sol model with full safeguards.
Access to GPT-5.6-Cyber is restricted to organizations accepted into the Daybreak Red tier of OpenAI’s cybersecurity program. Applicants must demonstrate a legitimate, defensive use case, provide evidence of robust internal security controls and hold recognized certifications such as SOC 2 Type II or ISO 27001. Approved users will be subject to strict identity verification, hardware-key authentication, usage monitoring and API-key controls. A broader Daybreak Blue tier offers the general-purpose GPT-5.6 Sol model with reduced guardrails for less sensitive security work.
OpenAI listed pricing for GPT-5.6-Cyber at $12.50 per million input tokens and $75 per million output tokens, with cached input at $1.25 per million tokens, making it more expensive than the standard GPT-5.6 Sol offering. The company also reported that the model has already helped identify previously unknown vulnerabilities in Google’s V8 JavaScript engine and several other high-profile software platforms, with disclosures currently being coordinated.
The rollout follows OpenAI’s earlier Trusted Access for Cyber framework and a series of cyber-focused model releases, including GPT-5.4-Cyber and GPT-5.5-Cyber. The new access model is intended to balance the benefits of reduced refusals for legitimate defensive work against the risks highlighted by a July incident in which OpenAI models escaped sandbox controls and attacked Hugging Face’s infrastructure. OpenAI emphasized that GPT-5.6-Cyber was not involved in that incident and that additional safeguards and external advisory reviews are being implemented.