OpenAI has begun a series of briefings for federal agencies, U.S. state governments, and the “Five Eyes” intelligence alliance countries regarding the capabilities of its new cyber model. According to a report on Axios, companies and government agencies worldwide are eager to get their hands on the most advanced artificial intelligence tools, which on one hand offer enormous advantages to cyber defenders, but on the other carry destructive potential if they fall into the hands of malicious hackers.
OpenAI held a special event in Washington for approximately 50 cybersecurity experts from across the federal government, where the capabilities of its new model, GPT-5.4-Cyber, were demonstrated. The model was recently launched as part of a tiered access program, and government officials are undergoing a rigorous vetting process similar to that of commercial clients. The company now appears to be adopting a dual-track approach: Distributing a public version with strict safeguards alongside a more permissive and advanced version released exclusively to authorized defense entities.
Chris Lehane, OpenAI’s Vice President of Global Affairs, noted at the event that this approach would also enable sensitive civilian bodies, such as local water utilities, to gain access to these advanced tools. Sasha Baker, the company’s Head of National Security Policy, added that OpenAI intends to establish partnerships with government offices and build channels for sharing threat intelligence across different sectors.
OpenAI’s offensive move comes immediately after the launch of competitor Anthropic’s Mythos model. The two companies are now competing for access to government agencies. However, while Anthropic has refrained from publicly releasing its model due to cybersecurity risks, its deployment within government has faced significant challenges. According to Axios, the Pentagon has labeled Anthropic a “supply chain risk” following a fierce dispute over AI safety, although it is important to note that the National Security Agency (NSA) is still evaluating the model.
As noted, OpenAI is conducting briefings with the “Five Eyes” countries - an intelligence alliance that includes the United States, Australia, Canada, New Zealand, and the United Kingdom - with the aim of approving their entry into the program. Most organizations that already have access to the new models are using them to identify security vulnerabilities in their internal systems. Since many government agencies contend with outdated computing systems that are difficult to secure, these AI tools are expected to help significantly accelerate the process of identifying critical weaknesses before it is too late.