OpenAI launches specialized model for cybersecurity tasks
OpenAI has released GPT-5.6-Cyber, a specialized model for advanced cybersecurity tasks. Access is restricted to approved organizations through a new program.
San Francisco – AI company OpenAI has launched GPT-5.6-Cyber, a specialized artificial intelligence model designed to enhance capabilities in advanced cybersecurity tasks. This includes vulnerability research and exploit development for approved defenders.
The GPT-5.6-Cyber model is a fine-tuned version of OpenAI's general-purpose GPT-5.6 Sol. It has been specifically trained to improve performance on complex cybersecurity scenarios, such as identifying zero-day vulnerabilities and constructing exploit chains. OpenAI has also focused on reducing refusals for high-risk "dual-use" requests that could be employed for either defensive or offensive purposes.
Internal benchmarks indicate GPT-5.6-Cyber achieved a 95% completion rate on advanced cybersecurity tasks. This significantly outperforms its predecessor, GPT-5.5-Cyber, which scored 57.3%, and the standard GPT-5.6 Sol model with its safeguards, which managed only 1.5%. Researchers describe the model as a large-scale effort to directly improve specialized cybersecurity expertise.
Access to GPT-5.6-Cyber is not broadly available. Organizations must be accepted into OpenAI's new Daybreak Red cybersecurity program tier to gain access. This program is intended for trusted defenders engaged in authorized, professional cyber work. A broader tier, Daybreak Blue, offers vetted users adjusted access to general models for defensive tasks.