The Pentagon-Anthropic Spat Is Good for China

 

 

“As the author of a book that Anthropic used without permission or payment to train its artificial-intelligence chatbot Claude, I can't say that the company's welfare tops my list of concerns. But as a historian and an American, I'm troubled by the Pentagon's behavior toward it.

 

Up to a point, the fight between the Pentagon and Anthropic is unremarkable. Contract negotiations break down all the time because the parties can't find mutually agreeable conditions. If all that is happening here were that the Pentagon won't buy under the conditions imposed by Anthropic, and Anthropic won't sell under the conditions imposed by the Pentagon, then there would be no cause for concern.

 

But that isn't all that's happening here. It's being reported that the Pentagon accepted the conditions that Anthropic wanted in a new contract with OpenAI -- which suggests that the Pentagon's real objection wasn't to the conditions, but to Anthropic. Buttressing that interpretation are public statements by Defense Secretary Pete Hegseth and President Trump castigating Anthropic for "wokeness." The Pentagon has gone beyond declining to buy from Anthropic by designating it a supply-chain risk, which prevents Anthropic from working as a subcontractor on other Pentagon contracts. Mr. Trump may soon issue an executive order directing all federal agencies to stop using Anthropic AI. These punitive moves have significant implications for the company's bottom line.

 

Hence the concern that the Trump administration is pursuing an agenda other than national security. It has created the appearance that it is choosing not to buy on technological merit while also punishing a world-leading American company in a race for AI that it claims the U.S. must win.

 

This isn't unprecedented -- and the precedent should worry Americans.

 

In the first decade of the 20th century, two brilliant English civilians, Arthur Pollen and Harold Isherwood, invented the world's first modern gunnery computer to aim the big guns of battleships. The most sophisticated computer of its day and decades ahead of its time, it represented a quantum improvement in the effectiveness of the Royal Navy's primary weapons system and the nation's premier instrument of combat power. It was the closest thing at the time to AI: a mission-critical technology capable of enabling smart, networked warfare and precision strikes. An observer who saw their computer described it as a "machine that uses intelligence."

 

Equipped initially only with ideas and not a working prototype, Pollen and Isherwood sought cooperation and support from the British Admiralty. So deeply did they impress the Royal Navy's leading gunnery officers that the Admiralty agreed in 1906 to an extraordinary development contract, in which the Admiralty provided funding and both parties undertook to preserve the invention's secrecy.

 

Pollen and Isherwood honored the contract, but the government didn't. Instead, the Admiralty fed information about their system to a naval officer named Frederick Dreyer, who was a capable gunnery officer but not the great inventor he fancied himself to be. (He was also a vindictive egotist, but fortunately there aren't any of those in the Trump administration.) Dreyer built a less capable but cheaper knock-off of the Pollen-Isherwood system. The Admiralty opted for the knock-off. It relied on Dreyer's ostensibly disinterested advice, which was backed by powerful superior officers with whom he had ingratiated himself, even though a majority of expert gunnery officers unequivocally opposed it.

 

This should have freed Pollen and Isherwood to sell abroad. But the Admiralty had no intention of letting them do so. So it resorted to smear tactics, made baseless accusations, and threatened them with prosecution under the Official Secrets Act if they tried. When Pollen and Isherwood called the Admiralty's bluff, it retaliated with the equivalent of designating their company a supply-chain risk: It struck their firm from its list of approved suppliers, thereby making them toxic to other defense contractors. It was a bad look for the navy of one of the world's two noisiest champions of private property and market freedom.

 

Pollen and Isherwood's resolution to export gave the navy of the other noisiest champion -- the U.S. -- the opportunity to pirate them too. The U.S. Navy, which wanted the best and thus had no interest in Dreyer's inferior system, funneled information about the Pollen-Isherwood system to an American startup. Headed by a much more talented engineer than Dreyer, the American firm produced a system as good as the English original.

 

There is a cautionary tale here for the U.S. today as it squares off against China. The Pollen-Isherwood story provides a lesson about the danger of decadence for the reigning hegemon in the face of a rising challenger. In the Pollen-Isherwood computer, Britain had a world-leading home-grown technology in a crucial sector. But the Admiralty took its eye off the ball. Instead of focusing on getting the best technology to serve the national interest, it corrupted its acquisition process. By contrast, the U.S. Navy had no agenda other than excellence in support of the national interest.

 

So there is precedent for the Pentagon's behavior toward Anthropic. But it isn't a precedent that bodes well for the U.S. in its competition with China over AI.

 

---

 

Ms. Epstein is a professor of history at Rutgers University and author, most recently, of "Analog Superpowers: How Twentieth-Century Technology Theft Built the National Security State."” [1]

 

1. The Pentagon-Anthropic Spat Is Good for China. Epstein, Katherine C.  Wall Street Journal, Eastern edition; New York, N.Y.. 11 Mar 2026: A15.