GPT-5.6-Cyber

GPT-5.6-Cyber is a model built by OpenAI specifically for security-testing tasks. Think of it like hiring a specialist technician who only inspects locks and wiring, instead of a general building manager who handles everything. It isn't meant for everyday conversation or writing—it's tuned to hunt for hidden flaws buried inside programs.

OpenAI introduced this model as part of expanding Daybreak, its cybersecurity initiative, and has said it has already been used to find previously unknown vulnerabilities in widely used open-source software, such as the core engine behind the Chrome browser. But the same ability that makes such a tool useful for defense can just as easily be turned toward offense—meaning the same model can serve as both shield and blade.

Notably, the model's release coincided with OpenAI disbanding the dedicated team that had been monitoring serious risks across its models more broadly. Expanding dual-use (offense-and-defense) security models while shrinking the team responsible for comprehensive risk oversight is a decision often cited as an example in debates over the direction of AI safety management.

Articles use it like: "OpenAI recently introduced GPT-5.6-Cyber, a cybersecurity-specific model, which found previously unknown vulnerabilities in things like Chrome's V8 engine." It's easy to miss that the model itself isn't the problem—the real point of controversy is that the expansion of dual-use defensive technology and the shrinking of risk-oversight teams happened at the same time.