CDR Salamander - Understand Anthropic vs. Department of War

Remember back in January the very confusing story about the conflict between Anthropic and the Department of War?

Well, at the end of last week, it broke back into the news.

A U.S. judge on Thursday permanently blocked the Pentagon’s “supply-chain risk” designation against Anthropic, ending a months-long standoff over military access to the company’s frontier AI models.

The dispute began early this year when the Pentagon pressed Anthropic to remove safety guardrails from Claude, a demand ​that escalated into a blacklisting and triggered lawsuits and federal agency bans.

I’m not going to go into a lot of the details that are floating out there. It gets real complicated, real fast—and in places conflicts with other reports. At best, most are second-hand or third-hand reports, circular reports, or the standard issue partisan positioning.

Firsthand reports from either side are few—especially about the specific details.

The above Reuters report is not that helpful, though it does supply a good timeline. This CNN report isn’t much better.

A federal judge ruled the Pentagon’s decision to label AI company Anthropic a ‘supply chain risk’ violated the law and ordered the designation be removed Thursday evening.

Judge Rita Lin wrote that while the military should have wide latitude to decide which companies to work with, its actions against Anthropic “constituted unlawful retaliation in violation of the First Amendment” and that the company “was denied the pre-deprivation process required under the Fifth Amendment.”

The Department of Defense’s quarrel with Anthropic began earlier this year after the company refused to remove internal safety guardrails that would prevent the military from using its Claude AI model in autonomous weapons and mass surveillance. Anthropic argued its models were not reliable enough to be used for such purposes.

Part of the problem is that most of the reporting usually focuses on very vague technical issues that are only quasi-germane to the issue at hand, or are play-by-play legal reporting. They are not useful from a military utility point of view.

Almost everyone here looks at things through a national security lens, so I can think of few people better to speak our language than Undersecretary of War for Research and Engineering (USW(R&E)) Emil Michael.

The good folks at the Ruthless Podcast have an exceptionally interesting interview with Michael that you should give it a full listen, but let’s focus on the Anthropic issue. Once I heard what had happened from his perspective, the issues became much clearer to me. When you look at it from a military operations perspective, the decision became unavoidable. You just can’t operate inside the constraints and restraints Anthropic insisted on.

If you just want to listen to the audio, go to the bottom of the post, I have the video of the interview there which provides an insider’s view of the Pentagon’s fight with Anthropic over AI.

If you’d rather see the transcript, here you go. Emphasis is mine.

Ruthless: …in your line of work, about why there was this massive uprising and problem with that situation. Anything you can—

Michael: Yeah. So this is because of the Biden executive order they’re talking about, where they tried to really constrain AI so they could control it. They had this rule that said if you use more than X teraflops of compute, you’re going to be regulated.

Ruthless: So crazy.

Michael: You’re like, okay, well then people are just going to use less compute and we’re going to lose to China, or whatever other ridiculous thing they had.But there was the chosen one company that helped them write that executive order.

Ruthless: Really?

Michael: Yes.

Ruthless; So they were trying to box in every EA person who were just like a menace, essentially. They didn’t bank on somebody like this showing up who actually knows. They’re like, ah shit, he knows the secret.

Michael: They were anointed as a winner. So where did they deploy first in the U.S. government? The Department of War and the intelligence community. What do they want to do? They want to control the system of government we have. They want to give it to machine god.I got the AI portfolio six months into the job. I open it up like, let me read the contracts here. I’m a former lawyer—or I went to law school, I didn’t practice law. Let me see what this says. They’re like, you can’t use AI for battlefield management. I was like, what does that mean? You can’t use AI to defend yourself from a missile attack. What? You can’t use our AI—25 pages of restrictions.

Indulge me as I break in here, as this really hit home. I almost got an eye-twitch flashing back to having to deal with all obscene “National Caveats” for troop contributing nations during my tour at ISAF in Kabul. Part of my responsibilities was to plan around these restrictions. “We have 2,500 ISAF forces in that district, but only 100 of them from nation X can be used for combat missions. Nations Y & Z that make up most of the force do not permit their forces to leave the gate.”

Back to the interview.

Ruthless: Imagine just writing that on a page with a straight face. With a straight face.

Michael: Yeah. And I was like, this can’t be right. Maybe someone didn’t read this contract. The Biden administration. Maybe someone didn’t read it.

Ruthless: It’s an onion contract.

Michael: So I called them in and I was like, okay guys, this has got to change. We’ll use it lawfully—all lawful purposes—but it’s artificial general intelligence. You don’t tell me where I can fly a plane to. We decide that when we buy a plane from Boeing or Lockheed. And they’re like, well, we think these are our moral values. We have our own constitution. It’s called the Claude constitution. It’s like, okay, but those aren’t corporate values. No, it’s our constitution. What about our Constitution?

Ruthless: We got one too.

Michael: Turns out there’s the Constitution, and everything else. And then they were so out to lunch. I’d say, okay, help me understand this. Let’s say I have a drone swarm coming against a military base. There are a thousand soldiers sleeping there. And I use AI to discriminate what’s a decoy and what’s not, and to point a laser at shooting it down. They’re like, huh, I think we could work on that. I was like, okay. How about a Chinese hypersonic missile going Mach 5 or above? I’ve got two minutes to take it down from space. So there’s a guy with a joystick who could do this.And they’re like, call me.I was like, well, call you when? Just give me a call. We’ll work it out. Did you miss the two-minute window?And they denied this afterwards, but I was like, I have 20 people in the room listening. There were former soldiers there who were about to lose it. Going crazy. And so this was the way it went.

You’re damn straight they were about to come unglued. Most of those guys were GWOT vets. We know stupid restrictions that make their way from the faculty lounge to your mission.

And then Secretary Hegseth is like, okay, you’ve got to stop messing around. We’re single-threaded on this one vendor at the time, so I need time to get other vendors in the door.He just calls them in. He says, I give you guys a week to work it out. And Thursday before the Friday deadline, they’re like, we’re pulling out. Trump’s a dictator. Elon’s a Stalin. Sam Altman’s Hitler. These guys are trying to destroy the world.

Ruthless: Partisan messaging.

Michael: Yes. There was no component of policy. I was like, we’re being—we could be unsafe. I was like, well, the law is the law. We have a democratic system. Laws get passed by Congress. The executive signs them. If I agree to use your system lawfully, what else do you want? They want to create their own law, their own standards.

There you go. DoW was spot on in keeping Anthropic away from national defense. That would be as if Northrup Grumman demanded that it gets a vote on how our B-2 bombers can be used.

Nope. The courts can say what they want, but they cannot force the U.S. military to use an untrusted vendor.

I assume this ruling from last week will run up the judicial chain.

Leave a comment

Share

View the full article