Pentagon Threatens to Terminate Cooperation with Anthropic

Reference News Network, February 18 According to a report by the Wall Street Journal website on February 14, informed sources revealed that the artificial intelligence (AI) model "Claude" from Anthropic was used to forcibly control the actions of Venezuelan leader Maduro by the U.S. military, highlighting the increasing application of AI models within the U.S. Department of Defense.

The forcible control of Maduro and his wife included bombings carried out at multiple locations in Caracas last month. Anthropic's usage guidelines prohibit the use of the "Claude" model to assist in violent actions, develop weapons, or conduct surveillance activities.

A spokesperson for Anthropic stated, "We cannot comment on whether the 'Claude' model or any other AI model was used for any specific action (classified or not). The use of the 'Claude' model—whether in the private sector or across the government—must comply with our usage policy. This policy outlines the ways in which the 'Claude' model can be deployed."

The U.S. Department of Defense declined to comment.

According to informed sources, the deployment of the "Claude" model was achieved through a collaboration between Anthropic and the data company Palantir Technologies. After the raid, an executive from Anthropic reportedly asked a colleague at Palantir Technologies whether the "Claude" model was used in the operation.

This newspaper previously reported that concerns from Anthropic about the U.S. Department of Defense's use of the "Claude" model have prompted government officials to consider canceling its contract worth up to $200 million. Informed sources revealed that company executives have publicly expressed concerns about the use of AI in autonomous lethal operations and domestic surveillance.

Anthropic is the first AI model developer to be used by the U.S. Department of Defense for classified operations. In this Venezuelan operation, it is possible that other AI tools were used for non-classified tasks.

Anthropic's CEO Dario Amodei and other executives have been publicly addressing the risks that AI models may pose to society. Amodei, differing from many other industry executives, has called for increased regulation and safeguards to prevent harm from AI.

This newspaper previously reported that during an event in January announcing the U.S. Department of Defense's collaboration with Musk's xAI company, Defense Secretary Pete Hegseth stated that the Department of Defense would not "adopt AI models that do not allow users to engage in warfare."

Amodei and other co-founders previously worked on the OpenAI platform, which recently joined a military-focused AI platform used by approximately 3 million users alongside Google's "Gemini" model. The OpenAI platform and the U.S. Department of Defense have indicated that a customized version of ChatGPT will be used for document analysis, report generation, and supporting research work.

Additionally, according to a report by Central News Agency on February 14, Axios News cited a government official on the 14th, indicating that the U.S. Department of Defense is considering terminating its partnership with the AI company Anthropic due to dissatisfaction with its insistence on limiting certain uses of its tools.

Reports indicate that the Pentagon is pressuring four AI companies to allow the military to use their tools for "all legitimate purposes," including weapon development, intelligence gathering, and battlefield operations, but Anthropic has not agreed to these conditions. Negotiations between the two parties have lasted for months, and the Pentagon has grown impatient with the situation. (Translated by Lin Chaohui)