The most important question nobody's asking about AI

Patel's entry point is a specific dispute between the Department of War and Anthropic, and he begins by conceding the government's side almost completely. The DoW has the right to refuse to use Anthropic's models because of the red lines. He even supplies the argument the government could have made, and calls it very reasonable: no one wants a world where the military becomes dependent on a private company for warfighting and can then be cut off if that company decides the terms have been crossed, especially given how ambiguous concepts like autonomous weapons or mass surveillance are.

He goes further, into a thought experiment that is meant to be uncomfortable for his own side: if he were Defense Secretary, he would probably refuse the deal. Imagine a future administration negotiating with SpaceX for Starlink access, and Elon Musk reserving the right to cancel if he determines the military is waging a war not authorized by Congress. The language sounds reasonable on its face, but a military cannot give a private company a kill switch on technology its operations rely on — especially with an acrimonious, low-trust contractor relationship, which is what Anthropic has with the current administration.

The distinction he then draws is the reason the post exists. If the government had simply said it would not do business with Anthropic, that would have been fine and would not have warranted writing anything. What happened instead is that the government threatened to destroy Anthropic as a private business because Anthropic refused to sell on terms the government dictates.

The mechanism is a Supply Chain Restriction: Amazon, Google, Nvidia and Palantir would have to ensure Claude touches none of their Pentagon work. Patel grants that Anthropic could survive the designation today. His argument is about trajectory — AI will not remain a party trick that can be switched off in a contractor's product, but will be woven into how everything is built, maintained and operated. The example is deliberately concrete: if the code for the AWS services the DoW uses is written by Claude, is that a supply chain risk? In a world of ubiquitous, powerful AI, it is not clear the big tech companies can cordon off Claude and keep working with the Pentagon.

That produces a question he says the Department probably has not thought through. If AI is that pervasive, then when forced to choose between their AI provider and a DoW contract representing a tiny fraction of revenue, most tech companies would drop the government, not the AI. Which leaves the plan as: coerce and threaten to destroy every company that will not comply on exactly the government's terms.