Pentagon reevaluates Anthropic partnership after Claude AI reportedly used in Maduro raid

The US military reportedly deployed Anthropic’s Claude artificial intelligence model during a high-stakes January operation to abduct Venezuelan President Nicolás Maduro, a move that has sparked a severe rift between the Pentagon and the safety-focused AI firm.

Following reports of the model’s involvement in the Caracas raid, a senior administration official told Axios that the Pentagon is now reevaluating its partnership with Anthropic. The official noted that Anthropic executives had inquired whether their software was used in the operation, signaling a potential disapproval that has caused “real concerns” within the Department of War regarding the reliability of the company as a long-term defense partner.

“Any company that jeopardizes the operational success of our warfighters on the ground is a company whose partnership we must reevaluate moving forward,” the official stated.

An Anthropic spokesperson flatly denied the allegations of direct collaboration on the mission. “Anthropic has not discussed the use of Claude for specific operations with the Department of War,” the spokesperson said. “Furthermore, we have not discussed this with any industry partners, including Palantir, beyond routine discussions regarding strictly technical matters.”

The incident underscores the growing friction between major AI laboratories and the US military. While companies seek to impose ethical guardrails on their tools, the Pentagon is increasingly demanding unrestricted access to real-time data processing capabilities, which are considered critical in the chaotic environments of active combat.

While Axios could not confirm the precise tactical role Claude played in Maduro’s abduction, the military has historically utilized the model for analyzing satellite imagery and intelligence. Sources familiar with the matter indicated that Claude was used not only during the planning phases but also throughout the active operation.

The Pentagon is currently pressuring AI giants to permit the use of their models for “all lawful purposes,” provided they comply with international and domestic law. Anthropic, which has branded itself as a leader in AI safety, remains in difficult negotiations with defense officials over its terms of service.

The company is specifically seeking guarantees that its technology will not be used for the mass surveillance of American citizens or the operation of fully autonomous weapons systems. A source familiar with the negotiations told Axios that Anthropic remains confident that the military has complied with its existing usage policies, which include these additional restrictions.

“We cannot comment on whether Claude, or any other AI model, was used for any specific operation, classified or otherwise,” an Anthropic spokesperson said. “The use of Claude in both the private sector and government agencies must adhere to our Usage Policies. We work closely with our partners to ensure compliance.”

Defense Secretary Pete Hegseth has prioritized AI integration, arguing that the US must accelerate its adoption of the technology to maintain a competitive edge over China. Sources familiar with the internal discussions suggest that top Pentagon officials are frustrated by Anthropic’s insistence on maintaining specific safety “red lines.”

Anthropic is currently one of several major model developers working with the Pentagon. Rivals such as OpenAI, Google, and xAI have already reached agreements allowing military users to access their models without many of the standard safety protocols applied to the general public.

However, Anthropic’s system is reportedly the only one currently integrated into the highly classified “air-gapped” systems where the military conducts its most sensitive work, ranging from weapons testing to active operational communications. This access is facilitated through a partnership with Palantir Technologies, a major defense contractor that incorporates Claude into its security products. It remains unclear if the Venezuelan deployment was directly tied to this Palantir integration.

While negotiations continue to bring models from OpenAI, Google, and xAI into these classified environments, the future of Anthropic’s $200 million contract remains uncertain as the Pentagon seeks to loosen the current restrictions on Claude’s deployment.