DoD, Anthropic now face legal, operational reckoning

Terry Gerton There are no shortage of newsworthy topics we could talk about, but one that’s certainly top of mind is the Pentagon’s dispute with Anthropic. And they’ve officially informed Anthropic that the company and its products are considered a supply chain risk, a status usually reserved for foreign companies. So, what does that mean now inside the federal procurement world?

Zach Prince Oh boy, so this is a complicated one. But for the last week, everyone who’s in this world has been scratching their heads because there have been these definitive statements that Anthropic is declared a security risk and everybody must terminate their contracts, but none of the statements that had been issued before had any legal force at all. They’re just tweets or declarations of the president or secretary of defense. None of those have the force of law. So, we’ve been trying to figure out what could be the bases for the government to issue orders, and how far could those orders go? It seems that they’ve picked a special DoD statute, 10 U.S.C. 3252, as a basis for their exclusion order. Under this statute, the secretary of Defense, Army, Navy or Air Force has to make a written determination that a source is a supply chain risk, which is specifically defined as an adversary, a risk that an adversary may sabotage, maliciously introduce unwanted function, or otherwise subvert a covered system. And a covered is specifically defined as a national security system. So think of systems, IT systems used in intelligence operations of one sort or another, or critical parts of major weapons systems, things like that. It’s not governmentwide. It is specifically relating to procurement of covered systems or things that go into the covered system. So assuming this is actually the basis of the order, or what we have been told is the order, it’s fairly narrow.

Terry Gerton And yet GSA took action to remove this from USAI.gov and provide guidance to all federal agencies to unwrap Anthropic from its AI tools.

Zach Prince Yeah, they sure have, and I don’t really understand why — well, why is because the president said so, right? But is that a legally recognizable basis to potentially terminate a contract for convenience, which is what I assume is going on in the background? Maybe and maybe not. Termination for convenience has historically been interpreted as an extremely broad right that government has to terminate the contract when it determines it’s in the government’s best interest. But it’s not totally unfettered. And this might be a case which finally defines the outer limits of when the government’s right becomes something that’s been done in bad faith.

Terry Gerton It seems like the dispute centers around Anthropic’s two red lines: no use of their AI tools for mass surveillance and no fully autonomous lethal weapons. And the Pentagon’s saying, “no, no not a problem, we will only use it for lawful purposes.” What do you think about Anthropic’s position that we really do want to have this written out in our contracts?

Zach Prince There’s a lot of analysis on this one. I get where Anthropic’s is coming from, because the Pentagon is saying this is already law — we, the Pentagon, cannot conduct domestic mass surveillance using AI tools and we cannot have fully autonomous weapon systems. Both appear to be true based on current law, but Anthropic’s worried that that might not always be true and they want to draw the red line now. Any technology company puts in its terms of use and its license limits on what you can use its services and its IT for. So this doesn’t strike me as unusual.

Terry Gerton I’m speaking with Zach Prince. He’s a partner at Haynes Boone. So Zach, let’s take the next step. Hypothetically, if Anthropic decides to fight this in court, what would that look like?

Zach Prince First you need an actual order, and my hope is that they don’t get there, either because the government backs down or because Congress does its job. Part of these orders requires notification to congressional committees. Congress knows that this is an outrage, and I don’t think that I am going too far in saying that this is an outrage. This is beyond the pale. I’ve never heard of an order being used anything like this way. It is obviously nothing to do with national security interests or protecting the interests of the United States against an adversarial attack. Exclusion orders, the only one that I’m aware of, was issued late last year for a Chinese military-backed cybersecurity company. So we’re talking about very narrow bases that this has ever been used. As far as I know, the government is still permitted to use the actual Chinese LLM model that’s out there. But there’s no exclusion order pertaining to DeepThink. So hopefully Congress just steps in and says, no, none of this. But as much as I think we’ve all hoped that Congress might step in in some of these cases, they haven’t. So it might be up to Anthropic to sue in federal district court, and they would seek injunction and usually I think that they would obviously get that temporary restraining order, an injunction against enforcement of the order, while the case plays out, because it’s so facially wrong. But sometimes when national security interests are claimed by the government, courts are loathe to take that first step until the case plays out. So we might have to see; the order might live for a bit before Anthropic ultimately prevails in court.

Terry Gerton Well, that timeline is going to be sort of nebulous. And in the meantime, it appears that DoD is using Claude and Anthropic’s tools in current operations, which has been a source of real concern from the Anthropic perspective. How is that going to play into all of this?

Zach Prince It really underscores the hypocrisy and the irony. DoD simultaneously last week was threatening to use the Defense Production Act to force Anthropic to sell its products — in other words, saying they’re critical for national security. And in fact, from all reports I’ve seen, they used Anthropic for targeting and other support for the Iranian effort. But also saying — you’re vital, but you’re a supply chain risk. So which one is it? These are not things that can live within the same reality.

Terry Gerton If DoD really does decide to unwind Claude from its tools, what does that kind of a timeline look like and how clean of an off-ramp can there be?

Zach Prince It’s very hard to say because a lot of the ways that they’re using Anthropic are in the classified realm. There are now potentially competitors that are allowed to be used in the classified realm. I know OpenAI came out with that agreement with DoD last week, so it’s possible they might be able to just quickly shift over to another large language model. But I’m skeptical. I think it will be very costly, and it will waste taxpayer resources for no discernible reason. But here we all are.

Terry Gerton You’ve been pretty clear that the situation is very uncertain and the legal ground is shaky. For companies who are government contractors and already using Claude, what does this mean for them?

Zach Prince We’ve been telling our clients for the last week that there’s no reason whatsoever to jump away from Anthropic and terminate contracts. But we’ve told folks it is prudent to be inventorying their use of Anthropic products. Come up with a list of all the places you’re using it, how you’re using it. Is it supporting government directly? Is it supporting DoD in particular? Is it a back-end, indirect support? And that way, when an order comes out, you can protect yourselves. If necessary, stopping use entirely, which I still have no idea of any legal basis the government could possibly direct that, or at least segregating it away from direct DoD contracts.

Terry Gerton Zach, everything that has preceded this point in this interview, you and I recorded on Friday, but this case is moving so quickly. There’s more real time news. So we’re inserting a real time update here. Tell us what’s new.

Zach Prince Sure. So as of today, March 9th, and there may be new things that happen before this even airs tomorrow, I don’t know, but Anthropic has filed suit in two separate fora, challenging orders that have been issued against it. One was, as we discussed under the 10 USC provision, that was filed in the DC Circuit, actually, which has immediate authority as opposed to usual appellate authority to review those types of orders. There was also an order that was apparently issued under FASCSA, it’s the Federal Acquisition Supply Chain Security Act of 2018. It provides for slightly broader exclusion provisions that Anthropic is challenging in the Northern District of California.

Terry Gerton Where do you think this goes next?

Zach Prince My guess is that they get an order staying any enforcement of this FASCSA order from the Northern District of California. I think this is obviously an unlawful order. It has no basis whatsoever in law or fact. I think that is what will happen. But who knows? DOJ might choose to immediately appeal if they get an injunction. Then we will have to see how it plays out.