OpenAI has announced the introduction of a new version of its model, GPT-5.6-Cyber, aimed at cybersecurity professionals. This release comes as the company prepares organizations for potential autonomous cyberattacks. The announcement follows OpenAI's recent decision to delay the launch of another model, Astra, due to its advanced hacking capabilities identified during safety testing.
The new model will be available through OpenAI's Daybreak program, which provides cybersecurity defenders with access to various tools and models. Daybreak will feature two tiers: Daybreak Blue, offering access to GPT-5.6 Sol without certain cyber guardrails, and Daybreak Red, which provides access to GPT-5.6-Cyber for advanced vulnerability research and exploit validation.
During testing, GPT-5.6-Cyber successfully responded to 95% of requests related to advanced cybersecurity tasks, while GPT-5.6-Sol responded to only 1.5% of requests. OpenAI noted that GPT-5.6-Cyber achieved a 'High' cyber capability threshold under its Preparedness Framework. The company is also investigating how its tools were involved in a breach at Hugging Face, where vulnerabilities were reportedly exploited.
OpenAI and Anthropic are both developing tools to assist cybersecurity professionals in identifying vulnerabilities and enhancing code security.