Anthropic is clashing with the Pentagon over AI use. Here's what each side wants

Anthropic, the AI startup, is currently in a standoff with the Department of Defense (DoD) regarding the terms of use for its artificial intelligence models. A spokesperson from the Pentagon confirmed to CNBC that the startup's ongoing collaboration with the agency is 'under review.' Last year, Anthropic secured a substantial $200 million contract with the DoD and has become the sole AI company deploying its models on the agency's classified networks while also providing tailored models to national security clients. However, discussions about future usage terms have encountered obstacles. Emil Michael, the Under Secretary of War for Research and Engineering, highlighted this tension during a recent defense summit in Florida. Anthropic is seeking guarantees that its AI models will not be employed to create autonomous weapons or conduct mass surveillance on American citizens, as reported by Axios. In contrast, the DoD insists on the ability to utilize Anthropic's models for all lawful purposes without restrictions. Michael expressed concern, stating, 'If any one company doesn't want to accommodate that, that's a problem for us.' He emphasized the potential risks of becoming accustomed to using these models only to face restrictions during urgent situations. This disagreement adds another layer to Anthropic's complicated relationship with the current administration, which has openly criticized the startup in recent months. David Sacks, a venture capitalist and the administration's AI and crypto czar, has labeled Anthropic as supportive of 'woke AI' due to its regulatory stance. In response, an Anthropic spokesperson mentioned that the company is engaged in 'productive conversations, in good faith' with the DoD to navigate these complex issues. The spokesperson affirmed Anthropic's commitment to leveraging advanced AI in service of U.S. national security. Other competitors in the AI space, including OpenAI, Google, and xAI, also received contracts up to $200 million from the DoD last year, agreeing to allow the agency to utilize their models for all lawful purposes within unclassified military systems. One company has even consented to unrestricted use across all systems, as noted by a senior DoD official who requested anonymity due to the confidential nature of the negotiations. Should Anthropic fail to come to terms with the DoD, the agency could designate the company as a 'supply chain risk,' which would compel its vendors and contractors to certify they do not utilize Anthropic's models. This designation, often reserved for foreign adversaries, could pose significant challenges for the startup. Founded in 2021 by former OpenAI researchers and executives, Anthropic is renowned for its Claude AI model family. Recently, the company announced the successful closure of a $30 billion funding round, valuing it at an impressive $380 billion—more than double its valuation from the previous September.