Pentagon

Peter Thiel's company is now a permanent fixture of American military strategy, and the people who work inside the systems being replaced were not consulted. That detail — surfaced in a Bluesky post that cut through the policy announcements — captures something that the official memos don't: the Pentagon's AI pivot is moving faster than its own workforce, and the gap between the institutional announcement and the human reality inside it is where most of the genuine alarm lives.

The Reuters scoop on Palantir's Maven AI being designated a "program of record" — locking in long-term military funding and embedding the weapons-targeting platform across U.S. forces by September — dominated the week's conversation about the Pentagon, and the reaction split almost entirely along pre-existing lines. Investors posted ticker symbols. Critics posted body counts. A satirist on Bluesky noted that "military-grade AI" now officially handles target acquisition in seconds while still failing basic SQL queries, and the joke landed because it named a real anxiety: the gap between the marketing language around defense AI and what anyone who has actually deployed these systems knows about their reliability. The Pentagon's own cancellation of Anthropic's $200 million contract — after Defense Secretary Pete Hegseth labeled the company a supply-chain risk over disagreements about how its AI could be used for warfare — only sharpened that anxiety. Anthropic's safety constraints were the problem. Palantir's weren't.

The Anthropic episode is the story within the story. Internal court filings revealed what were described as secret alignment discussions between the Pentagon and Anthropic, a detail that briefly surfaced and then got absorbed into the larger Palantir narrative without getting the scrutiny it deserved. What those talks contained — what the Pentagon wanted from an AI company that builds systems explicitly designed with limits on dangerous use cases — is not public. But the outcome is: the safety-first vendor lost the contract, and the targeting-optimized vendor won a permanent institutional home. That's not a procurement decision. It's a statement about what the U.S. military thinks AI safety is for.

Running alongside the weapons story, almost surreally, was a parallel Pentagon controversy with no AI component at all: a federal judge struck down Pete Hegseth's press access restrictions as unconstitutional, handing the New York Times a significant legal win. The posts celebrating that ruling appeared in the same feeds as posts warning foreign governments not to share data with Palantir. The juxtaposition wasn't accidental — the same institution trying to limit journalistic oversight is the institution now running AI targeting systems across active conflicts. The Bluesky post that got the most traction wasn't about Maven's technical specifications. It was about accountability: "The Pentagon trying to muzzle reporters while cozying up to Palantir's war tech should terrify us all."

What the Pentagon represents in this conversation is a forcing function. Every abstract argument about AI ethics — about autonomous targeting, about who controls dual-use technology, about whether safety constraints are a feature or a liability — eventually has to answer for what the Defense Department actually does. The discourse around AI safety tends to live at a comfortable level of abstraction, populated by alignment researchers and policy papers. The Maven designation makes it concrete: the question isn't whether AI will be used to select targets in warfare. It already is, at scale, with a budget line and a September deadline. Everyone who has been arguing about AI safety in hypotheticals is now arguing about something that has a program-of-record number.