The Pentagon's AI ultimatum to Anthropic, explained

The Pentagon wants basically the guardrails inside of Anthropic to be removed so that they have full access to the power of the model in order to defend our country. And Anthropic's very reason for existence was making sure that these kinds of guardrails are in place. The Pentagon essentially wants to be able to use Anthropic's AI model, CLAWD, however it sees fit. It doesn't want to be in a position where every single time it has to do an operation or do something of national security concerns, it has to check in with a company and make sure that it's following that company's specific safeguards. The problem with that is that Anthropic has two very clear red lines with its usage policies that it does not want crossed, even by the Pentagon, which are mass surveillance of Americans and autonomous weapons. When I read your story, I was sympathetic to Hegseth and the Department of War in that I feel like, Maria, if you choose as a tech company, listen, we're going to choose to work with the Department of War. It seems to me you have to say, yes, we grant all lawful use of our tech because I don't understand, Maria, how else it would work. Like if Hegseth, you know, he's okay, he's going to send Navy SEAL Team 6 into a battle zone and before those guys, you know, jump out of a C-130, they've got to check to make sure they're in compliance with like Anthropic's, you know, software agreement user terms. So a bit of context here that I think would be helpful. We know that Claude was used during the Maduro raid, right? So it was a very successful operation from the Pentagon. Claude was used. There were no hiccups there. And of course, the public doesn't know how these, how this technology is being used in classified settings, but we can surmise that there was no need for mass surveillance and there was no need for autonomous weapons. Anthropic's, you know, usage policies were not violated and yet the Pentagon was able to carry out a very successful operation. The second thing here with regard to the all lawful purposes standard that the Pentagon is trying to have, not just Anthropic, but all of the AI labs adhere to is that the law right now doesn't necessarily contemplate all of AI's applications. And so in the case of mass surveillance, for example, it is legal today for the government to collect data that is public. So for example, social media posts, concealed carry permits. If you attended a rally or a protest, voter registration rolls, the government can legally collect all of that data. And then you can imagine how injecting artificial intelligence can supercharge the in real time continuous analysis of that data to specifically target and surveil people. And that is kind of what is alarming, you know, not just Anthropic, but civil liberties groups and folks on Capitol Hill as well. The Pentagon has not specifically said that it wants to do mass surveillance or to use this technology for autonomous weapons. They actually view this as just a matter of doing business with the Pentagon and really important in an attack race with China. So that is something that they would argue on their side. In terms of how different it would be to mark a U.S. company as a supply chain risk, it's a very severe punishment. It's something that's usually reserved for foreign adversaries, companies coming from China, for example. The $200 million contract that Anthropic has with the Pentagon is a drop in the bucket to their $380 billion valuation. But what essentially blacklisting them would entail would be that any company that's doing business with the Pentagon would have to certify that they are no longer doing business with Anthropic. And that could be a bigger hit to the company. We reported yesterday that the Pentagon started the process of this supply chain risk designation. They reached out to Boeing and they reached out to Lockheed Martin to have them assess what their business with Anthropic looks like exactly. If the Pentagon does go ahead, they blacklist Anthropic, who potentially do you think stands to benefit? So you have three other AI labs, right, that are at play here. XAI, which already has a contract with the Pentagon for classified use in the all lawful purposes setting. You have Google and you have OpenAI. And we know that those two other companies are also accelerating their conversations with the Pentagon. And so they could potentially have a lot of business to gain here if Anthropic is kind of put to the side. We are monitoring, however, whether these two other companies have red lines of their own or any types of usage policies that they wouldn't want to see violated. The difficulty, though, with offboarding Anthropic is that they were the first to offer a specific model for the Pentagon. They're the only ones that are being used for classified purposes today. And so it would be kind of a pain to go through and get rid of them and replace them with a new AI lab. Do you have any sense of how this is all playing out in Silicon Valley? Is this seen as some new potential tension point, new fracture point between Silicon Valley and the Department of War? Or no, this is this is about Anthropic? I think that one thing we're looking out for right now is how employees might be reacting if this is something that they're comfortable with or that they might speak out on. We have seen employees speak out on previous situations, you know, regarding ICE, for example. I do think we can all see how fast this technology is moving. And so I do think that that having the leading player in this space sticking to its its core philosophy around guardrails and safety, I think, is really important. Tech companies now have this reputation as being the sort of these big evil empires, right? But in the beginning, Google's motto was don't be evil, like once upon a time when these guys started out as idealistic kids developing these technologies, they didn't necessarily feel this way. I think the biggest takeaway is that when you found a company like this at the very beginning, you better lay out what your point of view is. And then it makes it a lot easier to stick with it.