OpenAI has announced the introduction of a new AI model, GPT-5.4-Cyber, tailored specifically for defensive cybersecurity. This model is a “cyber-permissive” variant of GPT-5.4 and will not be available for public use.
The rollout is part of OpenAI’s preparation for the release of more advanced AI models expected later this year. The company aims to fine-tune its offerings to enhance defensive cybersecurity use cases, starting with the limited deployment of GPT-5.4-Cyber.
Access to GPT-5.4-Cyber is restricted to “the highest tier” of users, specifically those validated as cybersecurity defenders. Users must obtain Trusted Access for Cyber to utilize this model, which is designed with fewer capability restrictions for legitimate security work.
OpenAI stated that GPT-5.4-Cyber enables advanced defensive workflows, including binary reverse engineering. This functionality allows security professionals to assess compiled software for malware risks without needing source code access, thus improving cybersecurity efforts.
The iterative rollout targets vetted security vendors, organizations, and researchers due to the model’s enhanced capabilities. OpenAI is expanding its Trusted Access for Cyber initiative, which was launched earlier this year, to govern access to GPT-5.4-Cyber.
For individuals to gain Trusted Access for Cyber, they can verify their identity at chatgpt.com/cyber. Enterprises desiring access for their teams can follow procedures through their OpenAI representatives.
In addition to GPT-5.4-Cyber, OpenAI has recently upgraded its Pro plan for Codex users and released the GPT-5.4 mini and nano, which are positioned as its most capable small models to date. The company also advised users to update Mac applications, including ChatGPT and Codex, as a security precaution.
The release of GPT-5.4-Cyber represents a strategic enhancement to OpenAI’s focus on cybersecurity applications within its suite of AI products. The model aims to provide improved tools for professionals in the field navigating the complexities of cyber threats.
