OpenAI releases GPT-5.6-Cyber with 95% cybersecurity task completion

OpenAI has launched GPT-5.6-Cyber, a new AI model built specifically for cybersecurity professionals. The model reduces refusals on high-risk requests that could be used for both defense and attack, allowing approved defenders to research vulnerabilities and develop exploits. It’s a fine-tuned version of GPT-5.6 Sol, OpenAI’s latest general-purpose model released in June.

This release comes as cyberattacks grow more sophisticated and organizations struggle to keep up with rapidly evolving threats. By making it easier to develop exploits and test defenses, OpenAI is targeting the gap between what general AI models can do and what cybersecurity teams actually need. The timing also aligns with increasing demand for AI tools that can help prevent breaches before they happen.

On an internal benchmark measuring advanced cybersecurity tasks like exploit-chain development and privilege escalation, GPT-5.6-Cyber achieves a 95% completion rate. That’s a big jump from general models, which often refuse or fail on such requests. OpenAI says it trained the model to handle “dual-use” tasks—work that could be used for good or bad—with clearer boundaries and oversight. The company is positioning it as a tool for red teams, penetration testers, and security researchers working under strict ethical and legal guidelines.

What happens next could reshape how cybersecurity teams use AI. If the model meets its promise, other companies may follow, building similar specialized tools. But it also raises questions about misuse and oversight. OpenAI says it’s adding safeguards, including usage monitoring and restricted access. Still, the launch signals a shift: AI isn’t just for answering questions anymore—it’s being trained to actively probe weaknesses, and that changes the game entirely.