Anthropic, the AI developer behind the generative model Claude, has firmly denied allegations from the Department of Defense suggesting the company could sabotage its technology during military operations. In a recent court filing, Thiyagu Ramasamy, Anthropic’s head of public sector, stated that once the military is running the Claude model, the company has no ability to stop or manipulate it.
The accusations arose amid ongoing tensions between Anthropic and the Pentagon regarding the military application of AI. The defense establishment has labeled Anthropic a "supply-chain risk," which could restrict the use of its software, including by contractors, while other federal agencies have started to phase out the use of Claude.
In response, Anthropic has filed two lawsuits challenging the legality of the Pentagon’s actions and is seeking an emergency order to reverse its recent designation. A court hearing concerning one of these cases is set for March 24 in San Francisco.
Government lawyers have expressed that they cannot afford to risk military systems at critical moments, emphasizing the need for security regarding the use of Anthropic’s technology. They argue that Anthropic could theoretically interfere with military operations by discontinuing access or deploying detrimental updates if they disapprove of certain military applications.
Ramasamy refuted these claims, asserting that Anthropic has no backend access to modify or disable its models during such operations. He explained that any updates to the Claude system would necessitate approval from both the government and its cloud provider, which is not named but is understood to be Amazon Web Services. Importantly, he added that Anthropic lacks access to any data or commands input by military users.
Anthropic has advocated for a collaborative approach, stating it does not seek to exert control over military decision-making. In a contract proposal submitted on March 4, the company indicated a willingness to ensure its tool is not used for autonomous lethal actions. However, negotiations have not resulted in a resolution.
Currently, the Pentagon is reportedly taking steps to mitigate any perceived risks associated with Anthropic’s technology, including working with cloud service providers to limit Anthropic’s ability to make unilateral changes to the systems engaged in military operations.
