A race among the top AI companies to sell powerful models to the U.S. Defense Department is hotter than ever. No matter how the feud between Anthropic and the Pentagon ultimately plays out, the Pentagon is now more incentivized to contract with other tech firms. Likewise, whatever misgivings Anthropic had about working with the military have only grown.
Indeed, other companies are already taking steps to pick up the government-contracting mantle. Earlier this week, xAI reached an agreement with the Defense Department to operate on classified systems. And OpenAI is working on a Pentagon deal of its own. But winning over the Defense Department officials may not be enough. To actually become a go-to AI provider for the agency, their AI will need to catch up to Anthropic’s Claude large language model, which is widely liked within the military. And they’ll likely need to connect to Palantir’s technology.
Palantir, along with its partners, holds cloud security clearances that allow it to host highly sensitive military information and data. The company has also built a far more streamlined way of accessing data from across the DoD, and, presumably, data that would make any large language model far more useful to military officials. One former employee of the Defense Department’s Chief Digital and Artificial Intelligence Office tells Fast Company that Palantir has effectively “taken over the data lake problem” inside the Pentagon, consolidating raw and low-level data feeds and making them accessible through its platform.
“Everything runs through Palantir,” the former employee says. “They’re the 1,000 pound gorilla in this space.”
The dispute centers on the Pentagon’s demand that it be allowed to use Anthropic’s Claude model for “all lawful purposes,” while Anthropic has sought safeguards blocking uses for mass surveillance and autonomous weapons. After negotiations stalled this week, the Trump administration has reportedly was considering deeming the company a “supply chain risk,” which forces military contractors to ditch Anthropic models. On Friday afternoon, President Donald Trump posted on Truth Social that every agency was to immediately stop using all Anthropic products, with a six-month phase-out period. Shortly after, Defense Secretary Hegseth made the supply chain risk assessment official.
The Pentagon has already reached out to defense contractors to ask about their reliance on Anthropic. Palantir, notably, uses Anthropic models internally, one person tells Fast Company, and would possibly be impacted by the decision by the U.S. government to blacklist the AI firm’s technology.
Still, Anthropic has a real advantage in its integration with Palantir. “Since Claude is playing ball with [Palantir], it makes them more appealing than having to get Palantir to agree to share their stuff with OpenAI,” the former DoD employee says.
Even so, Claude’s agile technology remains a powerful draw. One recent government AI official says the LLM is so far ahead of its rivals that current and former government workers, (including those from the Defense Department) are sending memes about the standoff in at least one group chat.
Anthropic’s value to the Defense Department is also owed to the fact that its technology enriches the Maven Smart System, one former Palantir employee tells Fast Company. The Maven system—which has a long and controversial history—is an integrated platform that might help, for example, a military command team to access critical data that might be spread across the Defense Department. That data might include information about nearby munitions supply, or the number of soldiers that a military operation might be able to deploy.
Making these systems more interoperable makes it a lot easier to plan a military operation, the person said. While Anthropic could certainly try to independently sell its own system to the government, its technology is most useful to the government when integrated with a system like Maven. Palantir, the former Palantir employee added, wouldn’t be in a position to prevent OpenAI or Anthropic from connecting to something like Maven, but to be similarly useful as Anthropic those companies would likely want to enrich it, too.
From their understanding, it appears that Anthropic was early to gaining accreditation to work in these kinds of military systems, and other companies are still catching up.
Neither Palantir nor the Defense Department responded to a request for comment.