OpenAI held a Washington, D.C. event on April 21 where it demonstrated GPT-5.4-Cyber, its cybersecurity-focused model variant, to approximately 50 federal cyber defense practitioners, Axios reported. The briefings extended to state governments and the Five Eyes intelligence alliance (the U.S., U.K., Canada, Australia, and New Zealand), according to PYMNTS, citing a source familiar with the matter.
GPT-5.4-Cyber launched the prior week under OpenAI’s Trusted Access for Cyber (TACy) program, which provides a more permissive version of the model to vetted cybersecurity defenders, according to Axios.
Dual-Track Access
OpenAI is taking a two-tier approach to distribution. One version of GPT-5.4-Cyber ships broadly with standard safeguards. A second, more permissive variant goes to qualified cyber defenders through the Trusted Access program, according to Axios.
Chris Lehane, OpenAI’s Chief Global Affairs Officer, said during the event that this structure will let smaller organizations, including local water utilities, access advanced AI cybersecurity tools, according to PYMNTS.
Sasha Baker, who leads OpenAI’s national security policy, told attendees that the company hopes to work with government departments to prioritize the most critical use cases and share threat intelligence across sectors, PYMNTS reported.
The Competitive Context
The briefings came as rival Anthropic continues its own government outreach. Anthropic has been previewing its Mythos AI model to around 40 companies and organizations, including some within government, while holding off on wide release over safety concerns, according to PYMNTS.
The parallel efforts point to an emerging procurement model for government AI: specialized model variants with tiered access controls, rather than general-purpose models applied to security tasks. For agencies managing day-to-day cyber operations, the question is whether purpose-built models deliver meaningfully better threat detection and response than prompting general models with security context.
The Government Adoption Signal
The Five Eyes briefing is notable for its scope. Intelligence-sharing alliances have historically been cautious about commercial AI adoption, particularly for security-sensitive applications. OpenAI’s willingness to brief multinational partners suggests the company sees government cybersecurity as a long-term vertical, not a one-off contract.
The tiered access model, where the most capable version goes to highest-trust customers first, mirrors how defense contractors have historically distributed sensitive technology. Whether AI model access follows the same pattern at scale remains an open question.