5 Minutes
In Washington, people love a public breakup. They just don’t always mean it.
That’s the awkward lesson sitting underneath the Pentagon’s continuing use of Anthropic’s Claude—despite the loud, chest-thumping rhetoric that briefly made it sound like the U.S. military was ready to slam the door on the AI company for good.
Defense Secretary Pete Hegseth lit the fuse on X with a scorched-earth post aimed at what he framed as Anthropic’s “defective altruism.” The message was blunt: in his view, no Silicon Valley terms of service should outrank troop safety or battlefield readiness. Then came the threat that sounded like a hard cutoff—Anthropic would keep providing services for “no more than six months,” just long enough for the Department of Defense to switch to something “better” and, yes, “more patriotic.”
In the same breath, Hegseth painted Anthropic’s position as duplicity. Why? Because the company has drawn lines against certain future scenarios—mass surveillance at scale, or fully autonomous weapons—without objecting to what the Pentagon is doing today. That distinction matters. A lot. And it’s the part that tends to get lost once the word “betrayal” starts flying around.
Here’s the thing: when governments accuse a vendor of betrayal, you’d expect the relationship to end immediately. Yet reporting from the Wall Street Journal and Axios suggests the Pentagon kept right on moving with Claude in the workflow, even as the political theater played out.
According to the Journal, U.S. Central Command uses Claude in some capacity for intelligence assessments, target identification, and simulated battle scenarios. Those phrases are doing heavy lifting. “Some capacity” can range from mundane research support to sophisticated modeling that shapes how analysts think. But the direction of travel is unmistakable: large language models aren’t just office productivity tools anymore; they’re being threaded into the analytical machinery of modern defense.
The irony is that none of this contradicts what Anthropic has actually been saying. The company’s objections have largely focused on hypothetical future use cases—the nightmare versions of AI: pervasive surveillance states and machines making lethal decisions without meaningful human control. That’s different from assisting humans who are already tasked with intelligence work and operational planning, even if the ethical edges remain sharp.
While that debate was unfolding in the national security arena, Claude’s consumer profile was rising in a totally different one: the app store.
After former President Donald Trump reportedly took a swing at Anthropic’s culture, calling the team “leftwing nut jobs,” the Claude mobile app began climbing. Anthropic spokesperson Ryan Donegan told Gizmodo it hit number one in the U.S. App Store—an all-time high—overtaking ChatGPT as the most downloaded app. Whether the surge is fueled by politics, curiosity, or plain old product momentum is hard to pin down. Donegan also said daily signups have tripled over the past four months, which suggests the story is bigger than a single news cycle.
Meanwhile, OpenAI has been leaning into its own Pentagon narrative, promoting a deepening relationship that includes military applications in classified contexts. Sam Altman has suggested Anthropic may have wanted tighter operational control than OpenAI did—an interesting contrast, and a reminder that these companies are competing not just on model performance, but on governance philosophies and risk tolerance.
Still, it’s easy to get lost in the rhetoric of “future weapons” and “autonomous killbots,” because those are the images that stick. The present reality is both less cinematic and more consequential: AI systems that summarize, simulate, cross-reference, and model can shape decisions without ever pulling a trigger. That influence is precisely why the argument over guardrails isn’t academic.
And if anyone expected Anthropic to categorically refuse military work, the company’s own leadership has been pretty clear that’s not the plan. CEO Dario Amodei has indicated Anthropic is willing to work with the Pentagon as long as the relationship stays inside the company’s “red lines.” The debate, then, isn’t whether Claude belongs anywhere near defense. It’s who gets to define the boundaries—and how enforceable those boundaries remain once AI becomes another indispensable layer of the national security stack.
Leave a Comment