The U.S. Department of Defense is in talks with OpenAI and Anthropic to allow their AI tools to work not only in regular government systems but also in classified networks. The military is also pushing for the companies to remove some of the standard restrictions that usually apply to users.
The U.S. Department of Defense is in talks with OpenAI and Anthropic to allow their AI tools to work not only in regular government systems but also in classified networks. The military is also pushing for the companies to remove some of the standard restrictions that usually apply to users.
According to Reuters, the intention to deploy the most powerful generative models at all levels of access was spoken about by the Pentagon’s chief technologist Emil Michael during an event at the White House. The agency’s interlocutors clarified that this is about transferring commercial chatbots to «closed» networks, where sensitive data is processed. Such systems can be used for planning operations, analyzing intelligence and other tasks where an algorithm error is potentially very costly. When exactly and in what format the Pentagon plans to launch chatbots in secret circuits, is not yet disclosed.
A key dispute between the military and AI developers is who sets the boundaries. Companies are building safeguards into their models and requiring customers to adhere to usage policies. The Pentagon insists that it has the right to use commercial tools without «undue» restrictions, as long as everything complies with U.S. law. A separate risk is the ability of generative models to «invent» plausible but false answers, which can have deadly consequences in the face of classified information.
Separately, OpenAI this week signed an agreement to allow its tools to be used on the unclassified genai.mil platform. It is already deployed to more than 3 million US Department of Defense employees, and as part of the agreement, OpenAI agreed to relax some of the typical restrictions, although some safeguards remain.
According to the agency, Anthropic’s products can currently be used in classified systems through third-party contractors, and the government is still bound by the company’s policies. Anthropic, in particular, opposes the use of AI for autonomous weapons guidance and domestic surveillance, making the «unbundling» talks particularly contentious.
Previously, dev.ua wrote about how US Secretary of Defense Pete Hegseth stated that Grok from Elon Musk’s company would be used inside Pentagon networks along with Google’s generative tools.
