Anthropic Almost Got the Pentagon Contract Palantir Just Won

Palantir won the Pentagon's Maven Smart System contract. That's the headline. The story is what almost happened instead.

A court filing, surfaced by Reuters and picked up almost immediately on Hacker News, revealed that Anthropic had been within a week of signing a nearly identical deal before the Trump administration canceled it over what officials called an "ethics clash." Anthropic is now fighting a "supply chain risk" designation in federal court, with Microsoft backing the challenge. That backstory traveled faster than the Palantir news itself — and by the time it reached Bluesky, it had been stripped of its legal nuance and reframed as something more damaging: confirmation that the company most identified with AI safety had been quietly competing to become the infrastructure layer of American military AI.

The Bluesky reaction wasn't outrage so much as a particular kind of vindication. People who have spent years arguing that "responsible AI" is a brand position rather than a structural commitment found in the court filing exactly the evidence they'd been waiting for. The critique wasn't about Palantir's surveillance history or Pete Hegseth's role in the cancellation — those threads ran elsewhere, hotter and louder on X. On Bluesky, the conversation kept returning to a simpler and more uncomfortable point: if Anthropic and Palantir were competing for the same contract, the distinction the AI safety community has built its entire moral architecture around may be thinner than anyone wanted to admit.

What's missing from this moment is the usual counterargument. Military AI stories almost always generate a corrective layer of national security commentary — the adversarial framing, the North Korea angle, the "if not us, then who" logic that gives cautious observers something to hold onto. That argument is present in this news cycle, technically, but it's getting almost no purchase. The communities that usually perform measured optimism have gone quiet. Even the generalized dread on YouTube — autonomous weapons, killer robots, the science fiction vocabulary that normally keeps these discussions at a safe aesthetic distance — felt closer to the bone than usual this week.

The Anthropic-Pentagon near-miss is diagnostic precisely because it breaks the organizing fiction of AI ethics discourse: that there is a meaningful opposition between the safety-conscious and the reckless, and that this opposition maps onto which companies you should trust. A procurement filing doesn't sustain that story. The communities built around that opposition know it, and the silence where the counterarguments should be is more telling than anything anyone actually said.