The Pentagon has signed new agreements with several technology companies, including Nvidia, Microsoft, Reflection AI, and Amazon Web Services (AWS), to utilize their artificial intelligence (AI) tools and infrastructure on classified military networks. These deals mark an expansion of the US military’s AI capabilities following a breakdown in its relationship with Anthropic, a company that had previously provided its Claude AI model for use in sensitive operations.

The defence department announced the agreements as part of ongoing efforts to broaden its AI technology base. The new arrangements provide lawful operational access to proprietary AI models through firms like SpaceX, OpenAI, and Google, as well as open-source models supplied by Nvidia and Reflection AI. Microsoft and AWS are tasked with providing the necessary AI infrastructure. Emil Michael, chief technology officer at the Pentagon, emphasized the importance of open-source options, noting concerns about the infiltration of Chinese open-source models into American companies and highlighting the imperative for American alternatives.

The shift comes amid a contentious dispute between the Pentagon and Anthropic. The Claude model was until recently the only AI tool used in classified missions, including operations related to US actions against Iran and the capture of Venezuelan leader Nicolás Maduro. However, Anthropic’s CEO, Dario Amodei, refused to sign a new government deal unless the Pentagon guaranteed that the company’s technology would not be employed for mass domestic surveillance or in lethal autonomous weapons systems.

In late February, the Pentagon moved to designate Anthropic as a supply-chain risk, just one day before the US launched its military campaign in Iran, marking an unusual stance against a domestic tech firm. Defence Secretary Pete Hegseth has been highly critical of Anthropic’s leadership, describing it as “run by an ideological lunatic” and arguing that Anthropic should not influence military decision-making. President Donald Trump also issued a ban barring Anthropic from government contracts amid ongoing legal battles between the company and the defence department.

Despite the tensions, Amodei recently met with White House chief of staff Susie Wiles as the government explored access to Anthropic’s latest AI model, Mythos, which has demonstrated sophisticated capabilities in identifying and penetrating cyber defenses. Michael described the Mythos issue as a distinct national security concern, underscoring the need to strengthen network protections in light of the model’s ability to find and patch cyber vulnerabilities.

The Pentagon stated that these latest AI agreements support its goal of transforming the US military into an “AI-first fighting force,” enhancing warfighters’ decision-making advantage across multiple warfare domains as AI integration continues to accelerate under Secretary Hegseth’s leadership.