Anthropic is reportedly in a dispute with the Department of Defense and Defense Secretary Pete Hegseth over the use of its models.
Summary from Complicity Navigator:
Anthropic’s $200 million Pentagon contract is under review after the company resisted demands to allow unrestricted military use of its Claude AI model. According to reports, Anthropic has insisted on two limits: no mass surveillance of Americans, and no fully autonomous weapons. The Pentagon wants AI companies to permit “all lawful purposes” without restriction. OpenAI, Google, and xAI have reportedly shown more flexibility. The dispute came to a head after Claude was reportedly used in the military operation to capture Venezuela’s Nicolás Maduro, prompting Anthropic to ask its partner Palantir how the model had been deployed. Pentagon officials are now threatening to label Anthropic a “supply chain risk”—a designation typically reserved for foreign adversaries. Both sides say discussions are ongoing.
While I was at Microsoft, I had a small part in managing the company’s relationship with the Pentagon (during the Biden administration) and have an idea of what that relationship might be like for Anthropic. This is plainly a big story, but it’s likely even bigger than it seems.
For an enterprise technology company, the relationship with the Pentagon is absolutely critical. First, the obvious: the contracts tend to be enormous and relatively stable for the long term. For some tech companies, they are the anchor of the entire business. Beyond that, Pentagon contracts are a signal to other parts of the federal government, governments around the world, and major businesses everywhere of the quality of the product and the team behind it. Just like a degree from a selective college can open doors for a job-seeker, a Pentagon contract opens doors for the sales team, even for customers that have nothing to do with defense. You passed the American military’s vetting when it comes to things like reliability and security, so you’re a safe bet.
That preference is even more pronounced for any customer, whether another government agency or a company, that wants to work with the Pentagon, and all the companies that want to work with them, and so on. They want to use the same products and technical standards as DoD to make working together easier; if the Pentagon is using OpenAI models and tools built with them and you want to work with Pentagon, it’s better to use those same models and tools so you can assure the Pentagon procurement officers that everything will work together seamlessly. And, when you’re dealing with a vindictive and petty administration, why risk working with a company that they’ve blacklisted? In short: if Anthropic is frozen out by the Pentagon, countless other customers will have to consider dropping them as well.
So the $200 million that most of the reporting mentions is just the beginning of what’s at stake for Anthropic. There are billions more on the line.
And Anthropic is not a mature business with a long list of customers locked in by decades-long relationships, with piles of cash in reserve and an established brand that can weather this. They are a five-year-old startup fighting to win in the world’s most competitive and rapidly-evolving market. The economics of being a leading AI startup are staggering. Decisions have to be made years ahead under crippling uncertainty, on a scale of not merely billions but hundreds of billions and even trillions of dollars, fighting ferociously for engineering talent and compute capacity while realizing that, as CEO Dario Amodei recently explained, mistiming a prediction of the market by a year one way or another will create hundreds of billions in losses and wipe out the company.
All of that makes it so remarkable that Anthropic is not only challenging the Defense Department’s intended uses of its technology, but that it initiated these challenges. There is nothing easier than going along and rationalizing ethical breaches, particular when it’s your own government making the request. Consider how the other companies have handled themselves. Google, which has spent a quarter century making friendly logos in primary colors and whimsical Google Doodles and saying that their motto is “Don’t Be Evil,” appears to have been just fine with with its models powering mass surveillance of Americans (maybe unsurprising, since they’re also handing over personal information on critics of the administration to the Department of Homeland Security). OpenAI spent years setting up convoluted corporate structures so it could claim it was a non-profit intent on saving humanity. Many fell for it, but now the company is apparently fine with the surveillance (and is making contributions to keep the good times rolling).
When we think of political courage, we think of activists putting themselves in harm’s way or a politician doing the unpopular but right thing. That’s the stuff of books and movies, and it’s absolutely essential to democracy. Putting your business and your life’s work on the line for a principle counts too, even if it’s done from the comfort of a well-appointed conference room. If you’re not ready to call a major tech company or its wealthy executives courageous, consider this: if we want them to be better, we have to recognize and reward them when they are.
I hope Anthropic stands strong. I hope their customers and the world recognize the risk they’re taking and reward them with business and investment. I hope other companies take note.