OpenAI Briefs Federal Agencies and Five Eyes Allies on New Cyber AI Model
The company held a Washington event for roughly 50 cyber defense practitioners and is extending access to its GPT-5.4-Cyber model through a tiered program designed to keep advanced tools out of adversarial hands.
OpenAI has spent the past week briefing U.S. federal agencies, state governments, and Five Eyes intelligence partners on the capabilities of its latest cybersecurity AI product, GPT-5.4-Cyber, which was introduced through a tiered access initiative the previous week. The company hosted a Washington event on April 22 attended by roughly 50 cyber defense practitioners from a range of federal agencies, according to Axios, which first reported the meetings.
OpenAI is pursuing a two-pronged access strategy. One version of the model is being made broadly available with robust protective safeguards, while a second, less restricted version is reserved for verified defenders through what the company calls its Trusted Access for Cyber program.
Government applicants go through the same screening process as commercial entities. Chris Lehane, OpenAI's chief global affairs officer, told attendees the approach would allow organizations ranging from large federal agencies to local water utilities to access sophisticated AI tools for security purposes.
Sasha Baker, who leads national security policy at OpenAI, said the company is working with agencies to identify priority use cases and establish channels for sharing threat intelligence. Briefings with Five Eyes alliance members — the United States, Australia, Canada, New Zealand, and the United Kingdom — are ongoing this week, with those partners undergoing vetting before receiving model access.
The rollout follows Anthropic's restricted release of its own cybersecurity model, Mythos Preview. Anthropic has limited access to selected companies and at least two federal entities and has declined to make Mythos publicly available due to potential misuse risks.
The parallel approaches reflect growing government interest in advanced AI security tools and the delicate challenge of expanding access without enabling adversaries.
Read the original reporting at Axios.
