OpenAI Unveils GPT-5.4-Cyber: Federal Briefings Underway for New AI Cybersecurity Tool

OpenAI has been briefing federal agencies, state governments, and Five Eyes allies on the capabilities of its new cyber product over the past week, Axios has learned.

Why This Matters

Companies and agencies are eager to access the latest AI tools, which offer advanced cybersecurity capabilities that could significantly enhance defensive measures. However, these tools also pose risks, as malicious hackers could exploit them to advance their own tactics.

Key Developments

OpenAI’s GPT-5.4-Cyber Demo in D.C.

OpenAI hosted an event in Washington, D.C., on Tuesday for approximately 50 cyber defense practitioners across the federal government. The event showcased the capabilities of its new GPT-5.4-Cyber model, which was rolled out last week under a tiered access program.

Government applicants are undergoing the same vetting process as commercial customers seeking to join OpenAI’s Trusted Access for Cyber program, according to a source familiar with the matter. Attendees included officials from various government agencies and national security bodies, many of whom oversee daily cybersecurity operations.

Competition in AI Cybersecurity

OpenAI’s new cyber model follows Anthropic’s Mythos Preview, which was released to a limited audience of around 40 companies and organizations, including at least two federal entities. Anthropic withheld a public release of Mythos due to cyber risks.

OpenAI is adopting a dual-track approach: one version of its model will be widely available with robust safeguards, while a more cyber-permissive version will be accessible to defenders through the Trusted Access program.

Government and State Partnerships

At Tuesday’s event, Chris Lehane, OpenAI’s Chief Global Affairs Officer, emphasized that this approach would enable more organizations, such as local water utilities, to access advanced AI tools.

Sasha Baker, OpenAI’s head of national security policy, told attendees that the company aims to collaborate with government departments to prioritize critical use cases and establish channels for sharing threat intelligence across sectors. OpenAI is also working with state governments to provide them access to GPT-5.4-Cyber.

Five Eyes Briefings Underway

OpenAI has begun briefings with Five Eyes members this week to facilitate their vetting and access to the model. The intelligence-sharing alliance includes Australia, Canada, New Zealand, the U.K., and the U.S.

Challenges and Considerations

Anthropic’s rollout within the U.S. government has faced complications due to the Pentagon labeling the company as a “supply chain risk” following a dispute over AI safeguards. Despite this designation, the NSA is currently testing Mythos, as previously reported by Axios.

Industry Impact

Most organizations with access to OpenAI’s model and Anthropic’s Mythos Preview are using these tools to identify exploitable security flaws within their own systems. Many government agencies struggle with legacy systems that are difficult to secure and patch, making advanced AI tools a potential solution for accelerating flaw detection.