The relationship between the Pentagon and Anthropic has entered a serious crisis. The U.S. Department of Defense is considering ending its partnership with the company behind the Claude AI model because it refuses to remove all restrictions on the military use of its technology. What exactly is happening behind closed doors?
Dispute erupted after operation against Maduro
Tensions escalated following a recent military operation in which U.S. forces captured Venezuelan President Nicolás Maduro. Claude, Anthropic's AI model, was deployed during the operation through a partnership with data company Palantir. But then something unexpected happened. According to a senior government official, Anthropic's leadership contacted Palantir to ask whether Claude had actually been used in the operation. The tone of the conversation suggested that the company might have a problem with this – especially since shots were fired and people were injured during the operation.
Anthropic strongly denies this, however. A company spokesperson stated clearly that it had not discussed specific operations with the Pentagon or industry partners. The company maintains that its conversations focused exclusively on technical matters and fundamental limits in its usage policy.
"All lawful purposes"
The Pentagon is pressuring four leading AI companies – Anthropic, OpenAI, Google, and xAI – to allow the military to use their tools "for all lawful purposes". This includes even the most sensitive areas: weapons development, intelligence gathering, and combat operations. While OpenAI, Google, and xAI have already agreed to remove the standard restrictions applied to ordinary users, Anthropic is standing its ground. The company insists on two red lines:
- Mass surveillance of American citizens
- Fully autonomous weapons systems
The government official argues that these categories involve many gray areas and that the Pentagon cannot operate if it has to negotiate every individual use case with Anthropic, or if Claude unexpectedly blocks certain applications.
Anthropic signed a contract worth up to $200 million with the Pentagon last summer. Claude was also the first model the Pentagon deployed on its classified networks. This makes it a strategic partner, but also creates a problem if the partnership ends. "Everything is on the table," the government official acknowledged, including the possibility of scaling back the partnership or ending it entirely. "But a suitable replacement will need to be found if we decide that is the right response." The official also admitted that replacing Claude quickly would not be easy. "The other model companies are simply behind" when it comes to specialized government applications.
A cultural clash of values
According to the government source, Anthropic is the "most ideological" of all the AI labs when it comes to the technology's potential dangers. CEO Dario Amodei is known for his concerns about how AI could go wrong. But it is not just about the leadership. The company must also navigate internal unease among its engineers, who have reservations about working with the Pentagon. How can a balance be found between national security and ethical principles?
An Anthropic spokesperson emphasized that the company remains committed to national security work: "That is why we were the first advanced AI company to deploy our models on classified networks and the first to provide customized models for national security customers."
According to sources, the Pentagon is negotiating with OpenAI, Google, and xAI about moving into classified environments and is insisting on the "all lawful purposes" standard for both classified and unclassified use. One of the three has reportedly already agreed to these terms, while the other two are more flexible than Anthropic.
Will Anthropic yield to the pressure, or will the Pentagon find a new partner? This is not just about code and algorithms. It is about what kind of world we want to build.
Sources: axios.com and reuters.com
