April 14 - OpenAI announced GPT-5.4-Cyber, a tailored iteration of its latest flagship model designed specifically to support defensive cybersecurity work. The release follows the recent announcement by Anthropic of its frontier AI model Mythos on April 7, which is being deployed through Anthropic's controlled "Project Glasswing" initiative.
Anthropic's Project Glasswing permits select organizations to operate the unreleased Claude Mythos Preview model for defensive cybersecurity purposes and, according to the company, that model has identified "thousands" of major vulnerabilities across operating systems, web browsers and other software.
OpenAI said GPT-5.4-Cyber will be rolled out initially on a limited basis because the variant is designed with a more permissive approach toward sensitive cybersecurity tasks. Access will be granted to vetted security vendors, organizations and researchers during this early stage, the company said in a post on its website.
Expansion of Trusted Access for Cyber
In conjunction with the model announcement, OpenAI is expanding its Trusted Access for Cyber (TAC) program. The program, which OpenAI launched in February, will add new tiers of verification. According to the company, higher verification levels will unlock progressively more powerful capabilities for approved users.
Those approved at the highest tier will gain access to GPT-5.4-Cyber. The company noted that this model has fewer restrictions on sensitive cybersecurity tasks, explicitly citing activities such as vulnerability research and analysis.
Scope and initial user base
OpenAI said it intends to make TAC available to a broader set of defenders by expanding the program to thousands of verified individual defenders and to hundreds of teams that are responsible for protecting critical software. The company framed these steps as part of a controlled, tiered approach to distributing a model with relaxed constraints for certain cybersecurity uses.
What remains clear from OpenAI's announcement
- GPT-5.4-Cyber is a specialized variant of OpenAI's flagship model, fine-tuned for defensive cybersecurity work.
- The rollout will be limited at first, with access given to vetted security vendors, organizations and researchers.
- OpenAI is expanding its TAC program with new verification tiers; highest-tier users will receive access to GPT-5.4-Cyber and its fewer restrictions on tasks like vulnerability research and analysis.