OpenAI stepped up briefings over the past week with US federal agencies, state governments, and Five Eyes allies to showcase the capabilities of its new GPT-5.4 Cyber model, aiming to accelerate the use of AI in protecting digital infrastructure.
Details
The company held an event in Washington bringing together around 50 cybersecurity specialists, focusing on how the model can be deployed in defensive operations.
- OpenAI applies strict vetting for government access, similar to commercial clients, through its Trusted Access for Cyber program.
- Attendees included officials from government institutions and security agencies overseeing daily cyber operations.
- The company is offering two tracks: a broader version with tighter safeguards, and a more flexible one for defenders under the trusted access program.
- OpenAI is expanding partnerships with governments to define use priorities and build threat intelligence-sharing channels.
- It has already begun engaging with the Five Eyes alliance — the US, UK, Canada, Australia, and New Zealand — to grant access after vetting.
- It is also working with US state governments to scale adoption locally.
This comes alongside Anthropic’s release of its Methus Preview model, as both companies compete to work with governments on defining eligible users. Anthropic has held back a full public launch due to risks, limiting access to select institutions, including US federal entities.
Anthropic is also facing internal challenges within the US government after being flagged as a potential supply chain risk, even as the National Security Agency continues testing its model.
Organizations already using such models are deploying them to detect vulnerabilities in their systems, particularly as many agencies rely on legacy infrastructure that is difficult to secure — making these tools critical for accelerating the discovery of high-risk weaknesses.
What is Next?
Focus is now shifted to how these models will be distributed across government entities, and whether companies can balance defensive benefits with misuse risks.
