The Pentagon signs contracts with 8 AI vendors at once, with Anthropic dropping out from the sole provider.

According to Beating Monitoring, the Pentagon announced agreements with 8 technology companies, allowing their AI tools and infrastructure to be deployed on classified networks for the U.S. military, including Secret (IL6) and Top Secret (IL7) levels. The new signatories include Nvidia, Microsoft, AWS, and startup Reflection AI, along with previously signed companies xAI, OpenAI, Google, and the newly added Oracle on the same day.

Key terms of the agreement: all companies agree that the Pentagon will use their technology for “any lawful purpose.” Anthropic is not on the list. Previously, Claude was the only large model used by the Pentagon in classified environments, but CEO Dario Amodei refused to accept this standard, demanding a written government guarantee that it would not be used for large-scale domestic surveillance or lethal autonomous weapons. At the end of February, the Pentagon listed Anthropic as a supply chain risk, and Trump subsequently banned its participation in all government contracts. The two sides are currently in federal court. Defense Secretary Hegseth called Amodei a “ideological lunatic” on Capitol Hill.

However, The New York Times, citing sources, reported that at least some of the newly signed companies, while accepting the “any lawful purpose” standard, privately obtained assurances similar to those demanded by Anthropic, involving restrictions on autonomous drones and domestic surveillance.

Pentagon CTO Emil Michael stated that open-source models represent a new direction. Nvidia and Reflection AI provide open-source models, Microsoft and AWS provide infrastructure, and xAI, OpenAI, and Google provide closed-source models. Michael said Chinese open-source models have infiltrated many companies, and the U.S. needs alternatives.

Despite the Pentagon’s eagerness to replace Anthropic with OpenAI and Google, technical issues have arisen during the migration process, and intelligence analysts are still using Anthropic’s old models for daily work. Although Trump ordered the discontinuation of Anthropic before mid-year, Mythos has shown outstanding capabilities in cyber defense and offense, and The New York Times reports that this ban is almost certain to be revoked.