The Pentagon is considering ending its relationship with Anthropic over disagreements regarding limitations on the military’s use of its AI models, according to a senior administration official.
Standoff Over Military Applications
The dispute centers on the Pentagon’s push for access to AI tools for “all lawful purposes,” including sensitive areas like weapons development, intelligence gathering, and battlefield operations. Anthropic has not agreed to these terms, leading to months of demanding negotiations.
Anthropic maintains that its AI models should not be used for the mass surveillance of Americans or in the development of fully autonomous weaponry. The administration official stated there is considerable ambiguity around what would constitute those prohibited uses, and expressed concern about having to negotiate individual applications with Anthropic or having Claude unexpectedly block certain operations.
Recent Tensions
Tensions recently escalated following the military’s use of Anthropic’s Claude model in the operation to capture former Venezuelan President Nicolás Maduro, facilitated through Anthropic’s partnership with Palantir. According to the official, an Anthropic executive inquired with Palantir about whether Claude was used during the raid, implying potential disapproval due to the use of “kinetic fire” and resulting casualties.
An Anthropic spokesperson denied the characterization of that communication, stating the company had not discussed the use of Claude for specific operations with the Department of War or any industry partners outside of routine technical discussions. The spokesperson affirmed that Claude is used for intelligence purposes across the government, consistent with Anthropic’s Usage Policy.
The official described a broader “culture clash,” characterizing Anthropic as the most “ideological” of the AI labs regarding the potential dangers of the technology. However, the official also acknowledged that replacing Claude would be difficult, as other AI models are currently less advanced in specialized government applications.
OpenAI’s ChatGPT, Google’s Gemini, and xAI’s Grok are also used by the Pentagon in unclassified settings and have agreed to remove restrictions applied to general users. The Pentagon is currently negotiating with these companies to extend their use to classified networks, maintaining the “all lawful purposes” standard.
Frequently Asked Questions
What is at stake in the potential split between the Pentagon and Anthropic?
The Pentagon risks losing access to a leading AI model, Claude, which was the first frontier AI model brought into its classified networks. Anthropic risks losing a significant contract valued at up to $200 million.
What specific concerns does Anthropic have about military applications of its AI?
Anthropic insists on limitations regarding the use of its AI models for the mass surveillance of Americans and the development of fully autonomous weaponry.
What is the Pentagon’s position on Anthropic’s restrictions?
The Pentagon believes the restrictions are unworkable and seeks access to AI tools for “all lawful purposes,” arguing there is too much ambiguity in defining prohibited uses.
As the Pentagon and Anthropic navigate these disagreements, what level of ethical constraint should be expected from private companies partnering with the U.S. Military?