OpenAI is expanding cooperation with the U.S. government and key allies by rolling out GPT-5.4-Cyber, a new AI model focused on cybersecurity.
Axios reported on April 22 that, over the past week, OpenAI briefed U.S. federal and state agencies as well as Five Eyes partners—the U.S., U.K., Canada, Australia and New Zealand—on the capabilities of GPT-5.4-Cyber.
In Washington, D.C., OpenAI demonstrated the model for roughly 50 federal cybersecurity practitioners, showcasing potential operational uses.
GPT-5.4-Cyber is OpenAI’s answer to Anthropic’s Misos. Built on OpenAI’s latest flagship architecture, the model has been tuned for defense-oriented cyber missions. Officials say it should perform strongly at spotting system vulnerabilities, analyzing security threats and identifying insecure code.
Because of the risk of misuse, OpenAI is currently offering access only to vetted organizations and companies through a "Trusted Access" program.
OpenAI is also accepting applications for the cyber model via its website.
The company is pursuing a two-track approach: broadly distribute a version with strong safeguards while granting expanded capabilities to vetted security experts. OpenAI says this strategy will enable local public agencies and small-to-medium-sized organizations to leverage advanced security AI tools.
OpenAI is prioritizing collaboration with government partners to identify priority use cases and to build systems for sharing threat information across sectors. Security analysts expect AI to play an increasing role in public agencies that struggle to manage vulnerabilities in aging systems.
So far, OpenAI has promoted its cybersecurity model more aggressively than Anthropic. Anthropic's Misos remains restricted to roughly 40 "Project Wingglass" partners and a small number of governments, including the U.K.
Chan Park, reporter, cpark@aitimes.com
