OpenAI briefs US federal agencies and Five Eyes partners on GPT-5.4-Cyber model under Trusted Access program
OpenAI demoed its GPT-5.4-Cyber model to approximately 50 cyber-defense practitioners across the federal government at an event in Washington, D.C. on April 22, Axios first reported. OpenAI is beginning briefings with Five Eyes members this week (Australia, Canada, New Zealand, UK) to vet them for access to a more permissive version of the model under a
The Trusted Access tiering, corroborated by Axios and Benzinga, marks the first public instance of a commercial AI vendor constructing a formal Five Eyes access lane around its own product rather than entering through government procurement. The equities once managed through inter-service SIGINT-to-SIGINT sharing, offensive-cyber toolchains and vulnerability-identification pipelines, are migrating to commercial platforms where access is mediated by a vendor's vetting criteria, not IC channels. The framing may be primarily a marketing wrapper on a routine government-customer pipeline, with Axios's sourcing OpenAI-facing and the Five Eyes briefing not necessarily representing a formal intelligence-community arrangement. OpenAI racing Anthropic through the same IC sales channel is