Pentagon vs. Anthropic: AI release to be enforced with war law

In its dispute with Anthropic, the US Department of Defense has now allegedly issued an ultimatum until Friday, threatening the AI company with severe consequences if it does not meet the demands by then. This is reported by the US magazine Axios, citing informed sources after a meeting between US Secretary of Defense Pete Hegseth and Anthropic CEO Dario Amodei on Tuesday. If Anthropic does not grant the US military unrestricted access to its in-house AI, it will either be declared a risk to supply chains or forced to release a version for the military via the Defense Production Act (DPA), Axios quotes.

Law intended for production of war materials

While the first scenario is not new – even though classification as a supply chain risk was previously reserved for foreign adversaries – the alternative is no less extraordinary: The Defense Production Act is actually intended to enable the US government to order the production of war-critical goods. In recent times, however, it has also been used, among other things, in the context of the Corona pandemic to accelerate the production of medical devices and articles. Its use in a dispute over limits for AI systems would be a significant expansion of the law's scope. However, the AI industry has already been obligated to implement security measures when training its models with the help of the DPA.

Despite this escalation for the US government, Anthropic has so far shown no inclination to comply with the demands. The AI company does not want to allow the Department of Defense to use its AI technology for mass surveillance of the US population. Furthermore, weapons that fire without human intervention should not be developed with it. The department is nevertheless pushing for the release because it has no comparable alternative to Anthropic. “The only reason we're still talking to these people is we need them and we need them now,” the magazine quotes someone from the department. “The problem for these guys is they are that good,” it continues.

Anthropic is one of four AI companies to which the Pentagon pledged up to 200 million US dollars each in the summer for the development of agentic AI workflows. The US Department of Defense is now threatening to revoke this contract if the company does not comply with the demands. This would be significantly less problematic for Anthropic than the escalation. Classification as a supply chain risk would force other companies to choose between doing business with the US military or with Anthropic. This decision would likely not be difficult. At the same time, however, it has also been stated by the US Department of Defense that Anthropic's AI would be enormously difficult to remove from its systems.

(mho)