The Pentagon dropped the bomb a few hours ago. Anthropic has until this Friday at 5:01 p.m. to agree to have its AI models used for whatever the Pentagon deems appropriate. It is something that is turning the conversation on its head upside down. ethical use of AIbut more important than that, it puts on the table a very clear intention on the part of the United States: that if they consider that all resources must be used in matters of national defense, all resources belong to the State.
And that’s where artificial intelligence from private companies comes in.
In short. Dario Amodei is the CEO of Anthropic. The name may sound more familiar to you. Claudeone of his models. These are very versatile, but they have a red line very clear: They cannot be used for mass surveillance of US citizens (be careful with the nuance). Nor for the development of weapons or the use of autonomous weapons controlled entirely by AI. And this is something that conflicts with what the Pentagon wants: to use Anthropic models without restrictions.
Why Anthropic, you might ask? Well, because this company offered its tool to the Government for the symbolic price of one dollar, which earned them a contract of 200,000,000 dollars. The move did not go badly, and little by little their models were integrated with those of Palantir. Now, the US wants to have an ‘unleashed’ AI, but as my colleague Javier Lacort explainsthe laws and ethics of the United States Armed Forces are based on the fact that a soldier can and must disobey a manifestly illegal order.
An AI can do whatever you want it to do.
A lever from 1950. A 200 million dollar contract is juicy, but seeing the nonsense that is continually invested in AIit’s pocket change. Anthropic would have it as easy as refusing and clearing the way for someone else to take his place in the Pentagon, right? Well… it’s not that simple. And it is not for two reasons.
The first is that it is already highly integrated into all the systems, processes and services that the Department of Defense uses on a daily basis. It is too big a ship for it to change course. He “knows” too much about the interiors of those systems that must be highly secret. But the most important thing is that the United States has a lever to take over whatever it wants. The only thing you have to argue is that it is due to a need for national security.
In 1950, during the Korean War, the Defense Production Acta law that grants the president powers to secure supplies necessary for national defense. This, for example, gives the Government power to require companies to prioritize contracts with the State to allocate materials, services and facilities. Also allow aid to expand productive capacity and, ultimately, put the largest companies in the country to work depending on what is needed.
Too powerful to be private. This document has been expanded and expanded over time, since the needs of 1950 are not the same as those of later decades, and in 2023 the Executive Order 14,110. It is one that forces the owners of the most powerful AI models to notify the Government when they train systems that may pose a risk to national security, the economy or public health. After him, other drafts have been signed expanding the demands on AI, but the truth is that the Pentagon does not need to publish anything new to pressure Anthropic.
Under Title I (ordering that certain products or services be provided to the Government), the Pentagon could take over Anthropic’s AI to carry out its designs. The support is what we have already mentioned: that “national defense.” And, in practice, it is defined that, if the company does not want its AI to do certain things because of that kind of moral code, if it becomes the property of the Pentagon, the ethical barrier can be eliminated.
Disciplinary punishment for the “AI woke”. And now, the big question. Can Anthropic say something like “look, you keep the 200 million and each to their own”? Well no. Or not in such a simple way. I have already commented that Claude is too deep into the Pentagon systems (to the point that he was used in the operation to capture Nicolás Maduro) and “knows” too much. She’s very integrated into the Pentagon’s classified systems, and as we can guess, they weren’t going to let her go that easily.
And that’s where the figure of Pete Hegseth comes in. He is the Secretary of Defense, and he has qualified that Anthropic moral standard like that of an “AI woke.” Yes… Because Google, OpenAI and xAI allow that the Government uses its models in any “legal” scenario (for the Government, of course), but not Anthropic, and in addition to the lever of the Defense Production Act, the United States can harshly punish the company.
If they do not “give in” to their requests, they could identify Anthropic as a company that puts the country’s supply chain at risk. It would be cLike the Huawei veto a few years ago, but to a national company. In practice, US companies would not be able to work with Anthropic. That is to say, the country has a couple of good “arguments” for Anthropic to hand over that moral bar.
Learning. It won’t be that long until 5:01 p.m. on Friday, February 27, when we will know if Anthropic agrees by hook or by crook to the Pentagon’s “requests.” What is clear is that this is a lesson. Not so much for American companies that, in the end, operate under the umbrella of the Defense Production Act, but for those of the rest of the world. Specifically, the European ones. And I’m thinking about the space race.
Europe, like China, Russia and the United States itself, has embarked on the new space race. They are going at a different pace because, until not so long ago, they had the United States as that partner and ally that would always be there. The tariffs and the ukrainian waramong other things, have broken that trust and Europe has realized that You have to get your act together on many fronts..
And one of them is who you trust. For example, SpaceX has become a partner in satellite communication and space shipments, but SpaceX is no longer just a rocket company: is a company that owns xAI. And, as I mentioned, xAI has ties to the US Department of Defense. But also SpaceX is X -Twitter-a tremendous database of European citizens. And there is the other reading: foreign countries they cannot depend in critical systems from suppliers that operate under the US umbrella… and the Defense Production Act.
GIPHY App Key not set. Please check settings