OpenAI expands cybersecurity program with GPT-5.4-Cyber model

OpenAI will be expanding its cybersecurity efforts by increasing the number of verified defenders served by its Trusted Access for Cyber (TAC) program into the thousands, with hundreds more security teams to follow.

This move aims to address the challenge of defenders and attackers increasingly using AI, accelerating the pace and complexity of cyber threats.

At the heart of the release is the launch of the GPT-5.4-Cyber, a variation of the existing model that has been tailored explicitly for use in defensive cybersecurity contexts.

The model is designed to be ‘cyber-permissive’ in that it allows conducting vulnerability assessments and performing binary reverse engineering to understand software vulnerabilities without access to its source code.

It is intended for use in a restricted fashion by select individuals and firms.

This effort aligns with OpenAI’s overarching cybersecurity approach, which comprises three basic tenets: democratization of defense mechanisms, deployment through iterations for ...