OpenAI Launches GPT-5.6-Cyber for Security Research

OpenAI has introduced GPT-5.6-Cyber, a specialized artificial intelligence model built for vulnerability research, penetration testing and incident response. The company unveiled the model this week as part of a push into cybersecurity-specific tools.

Built on the company’s GPT-5.6 Sol architecture, the new model is trained to improve performance on tasks such as finding zero-day vulnerabilities and developing exploit chains. OpenAI said it has also reduced refusals for certain higher-risk, dual-use cyber tasks that the model would normally decline.

The company is making GPT-5.6-Cyber available through a new access tier called Daybreak Red, designed for firms conducting authorized vulnerability research, exploit validation and security testing. Access is restricted rather than opened to the general public, a decision OpenAI has tied to the model’s dual-use capabilities.

The move places OpenAI in more direct competition with specialized cybersecurity AI vendors and comes as offensive security researchers increasingly use large language models to speed up vulnerability discovery. Security teams have used earlier general-purpose models for similar tasks, but with more built-in resistance to exploit-generation requests.

OpenAI has not disclosed pricing for Daybreak Red access or named specific firms granted early use of the cybersecurity model. The company has previously said it screens applicants for programs involving reduced safety refusals, though it has not detailed the vetting process for this release.

The launch adds to a broader industry debate over how AI companies should balance genuine security research needs against the risk that more capable exploit-generation tools could be misused. Rivals including Anthropic and Google have taken more cautious public stances on similar dual-use capabilities, keeping most offensive security features behind tighter internal review.

Whether GPT-5.6-Cyber changes that balance will likely depend on how OpenAI manages access to Daybreak Red as demand grows. Cybersecurity researchers have welcomed more capable defensive tools while warning that any model trained to find exploits faster inevitably shortens the window defenders have to patch newly discovered flaws.