OpenAI introduced a model called GPT-5.5-Cyber on Monday as part of a set of announcements intended to deepen its work with cybersecurity companies and researchers. The release drew relatively little public or political attention compared with the high-profile export controversy surrounding Anthropic.
Performance and access
According to OpenAI, GPT-5.5-Cyber scored 85.6% on CyberGym, an internal benchmark that measures whether an AI agent can reproduce known software vulnerabilities. By comparison, Mythos 5 scored 83.8% on the same evaluation.
An OpenAI spokesperson told Axios that GPT-5.5-Cyber does not include cyber safeguards, so it can be used for red teaming and penetration testing, but access is restricted to defenders who are in the highest tier of its Trusted Access for Cyber Program.
Policy and regulatory questions
The relatively low-profile release raises questions about what specifically prompted the Trump administration's concerns about Anthropic's Fable 5 and Mythos 5, since a similarly capable system from OpenAI has been made available without comparable export restrictions or public political backlash.
The White House did not respond to requests for comment. Reports also suggest that personality clashes between Anthropic and the Trump administration may have contributed to the decision to impose export directives.
International partnerships and diplomatic context
OpenAI said it has expanded partnerships with organizations in Australia, Canada, France, Germany, Japan, Poland, South Korea and the European Union. Meanwhile, negotiating access to Mythos was a significant talking point at last week's G7 Summit.
Why this matters
The episode highlights the delicate balance around regulating and controlling advanced AI tools for cybersecurity: systems that demonstrate similar technical capabilities can prompt different political and regulatory reactions, affecting who can access these tools and under what conditions.
Editor's note: this story was updated with a comment from OpenAI.
