Washington Expands Its AI Allies, Anthropic Under Scrutiny

The Pentagon Expands Its List of AI Suppliers

The U.S. administration has recently decided to expand its list of artificial intelligence suppliers by adding four new companies. Microsoft, Amazon, Nvidia, and Reflection AI have been selected to provide their technologies for classified operations. However, Reflection AI has not yet released a publicly accessible model. These companies join OpenAI, xAI, and Google, which were already partners of the Department of Defense for uses qualified as "any lawful purpose."

Tensions with Anthropic AI

Anthropic AI, which was previously a key player, now finds itself at odds with the U.S. government. CEO Darius Amodei has expressed concerns regarding the use of Anthropic's technology for civilian population surveillance and the development of autonomous weapons. In response to these concerns, the Pentagon canceled a $200 million contract with Anthropic. This decision has led the company to file a lawsuit to recover lost revenue. The Trump administration had previously labeled Anthropic as a "supply chain risk," a first for a U.S.-based company. Additionally, government sources have described Anthropic as a "woke" company.

Goal: Avoiding Technological Lock-In

In its statement, the Pentagon emphasized its intention to build an architecture that prevents lock-in with AI suppliers, thereby ensuring long-term flexibility for the joint force. The new technologies will provide fighters with the necessary tools to act confidently and protect the nation against any threat. AIs will be used for use cases of impact levels six, which concern secret data, and seven, for the most highly classified materials. This contributes to creating an "AI-driven combat force."

Current Use of AI by the Pentagon

Currently, the Pentagon's use of generative AI is largely limited to unclassified tasks performed within the various defense departments. These tasks include document drafting and research. The new suppliers will also assist defense forces in streamlining data synthesis, enhancing situational awareness, and improving decision-making for fighters in complex operational environments. It is unclear whether these descriptions include domestic deployments within U.S. borders.

Anthropic's Persistent Role Despite Tensions

Despite the current tensions, Anthropic's Mythos model continues to be used by the NSA for its cyber warfare and defense capabilities. Globally, the Mythos model is currently being evaluated by 40 organizations, of which only 12 have been named, with the UK's MI5 and the U.S. NSA among the remaining 28. Anthropic's Claude AI had been used on classified materials as part of Palantir's Maven tool. According to Axios, the U.S. administration may reconsider its most recent public stance regarding Anthropic. A source at the White House indicated that the administration is looking for ways to "save face and reintegrate them," suggesting that Anthropic's Claude coding model is still being used by U.S. government security organizations.