OpenAI offers a new cyber weapon for US intelligence and its allies

Last Tuesday, OpenAI held a special event in Washington for about 50 cyber defense experts from all over the federal government, where the capabilities of the new model, GPT-5.4-Cyber, were demonstrated. The model was launched last week as part of a phased access program, and government officials go through a strict screening process identical to that of commercial customers. It seems that the company is now taking a two-track approach: distributing a public version with strict security measures, alongside the release of a more permissive and advanced version for qualified security agents only.

This week, as mentioned, OpenAI will begin briefings for the countries of the “Five Eyes”, an intelligence alliance that includes the USA, Australia, Canada, New Zealand and the United Kingdom, with the aim of confirming their entry into the program. Most of the organizations that already have access to the new models use them to locate security holes in their internal systems. Since many government agencies are dealing with outdated computer systems that are difficult to secure, the artificial intelligence tools are expected to help them significantly speed up the process Identifying the critical weaknesses before it is too late.