The Pentagon has threatened to designate Claude’s maker Anthropic as a “supply chain risk,” which would not only bar Anthropic from government contracts but also force Pentagon contractors to cut ties with the company. This is a devastating penalty typically imposed on foreign adversaries, such as Chinese telecom company Huawei and Russian cybersecurity firm Kaspersky Lab. Anthropic’s attackers insist that military uses of their artificial intelligence (AI) adhere to Anthropic’s two red lines: no mass surveillance of American citizens and no fully autonomous weapons. In response, a senior Pentagon official told Axios that the Pentagon is “definitely going to pay a price.”
Both sides have real arguments here, but the way the government has imposed its position is disproportionate to say the least. But the deeper question is not who is right in this negotiation. That means negotiations are happening at all. The terms governing how the military uses this century’s most innovative technology are set by bilateral negotiations between the secretary of defense and the CEOs of startup companies, with no democratic input or permanent constraints. Congress should set these rules. And it should be done quickly.
Both sides have a point.
A system built on private property and the rule of law allows businesses to choose with whom they do business and on what terms. Anthropic is under no obligation to sell products to the military without reservation. And the company’s red line is by no means frivolous. These echo concerns that CEO Dario Amodei has consistently expressed publicly, most recently in a January essay arguing that democracies should use AI for national defense “except for those that approach authoritarian adversaries.”
But democratic governance requires that the military, the public side of the military-industrial complex, be held accountable for how it uses its tools. We don’t want Lockheed Martin to sell the F-35 to the military and tell the Pentagon what missions it can fly. What I have argued in the context of government surveillance also applies to military matters. Decisions about government power should be made through a democratic process, not by private companies unilaterally constraining government through product design.
In principle, these two points are resolved fairly neatly. Businesses should not be forced to participate in work they deem objectionable, but that means they could lose government business. Humans can maintain that red line. The Department of Defense could find another vendor. Neither side can conscript the other.
Easy case and hard case
In reality, things aren’t always so pretty. Given the Trump administration’s actions, it’s easy to side with Anthropic. That red line is certainly a reasonable starting point for thinking about the responsible use of AI. And the Pentagon isn’t just looking for flexibility. It demands the right to use AI for “all lawful purposes” without restriction. This may sound reasonable until you consider that existing surveillance laws were enacted long before AI could monitor millions of people simultaneously. “Legal” covers much more ground than it used to, and I don’t trust this administration to stay within such generous boundaries.
The Pentagon’s supply chain risk threat further exacerbates the situation. If the Pentagon just wants another contractor, that’s fine. However, this designation would amount to a secondary boycott, prohibiting government contractors from using Anthropic as a subcontractor and, in some cases, requiring them to discontinue Anthropic’s services altogether. This is a lot of pressure. Eight of the 10 largest companies in the United States reportedly use Anthropic’s products.
It is also unclear whether the designation is legal. The relevant legislation (10 USC § 3252 and the Federal Acquisition Supply Chain Security Act (FASCSA)) was designed to target foreign adversaries that could compromise defense technology, rather than domestic companies that maintain contractual usage restrictions. The law covers acts such as “sabotage,” “malicious introduction of undesired features,” and “subversion,” i.e., hostile acts aimed at undermining the integrity of a system. Companies that openly restrict certain uses of their products through licensing agreements are doing something entirely different. So far, the only FASCSA order has been against Acronis AG, a Swiss cybersecurity company with reported ties to Russia. Anthropic is not Acronis.
And this designation is strategically counterproductive. Humanity has so far more It has been more willing than most AI companies to work with the military. It was the first Frontier Lab to be deployed on a classified network, and Claude was reportedly used in a military operation to capture Venezuelan President Nicolás Maduro. Punishing one company for showing up sends exactly the wrong signal and threatens to cripple one of America’s national AI champions. As Dean Ball, who worked in the Trump White House and was the lead author of the administration’s AI action plan, says, supply chain risk designations are not necessary when “cheaper options are on the table.”
But it’s easy to side with Anthropic on this regime. This difficult case requires thinking beyond President Trump and Secretary of Defense Pete Hegseth. Imagine a different, more normal administration in 2029, whether Democratic or Republican. Shouldn’t future presidents be able to withdraw from xAI’s technology if it refuses to assist in asylum procedures or civil rights enforcement?An example of that flexibility is concrete. As JB Branch explained in Lawfare, despite Grok’s “documented history of biased, misleading, anti-Semitic, and harmful output,” the current administration has deployed xAI’s Grok across the Pentagon’s classified networks and at the Department of Energy’s Lawrence Livermore National Laboratory. A future administration should absolutely be able to end it.
And even Anthropic’s own redlines become more complicated when you remove Trump from the variables. How much automated surveillance is actually appropriate? AI-assisted analysis of public information may be just what intelligence agencies need to identify real threats. Could autonomous weapons reduce both civilian and military casualties in some scenarios? Reasonable people object to this, and our system requires Congress to make such demands.
Parliament’s appeal
The rules for military AI should not depend on the ethical commitments of a CEO who happens to be in office or the political preferences of a secretary of defense who happens to be in office. Parliament should decide. But while Congress has imposed some restrictive reporting requirements and governance structures, it has not set substantive rules about which AI applications the military can and cannot pursue, even as AI threatens to dramatically expand executive powers and enable a national security apparatus that resists large-scale enforcement and oversight.
There are also practical reasons why Congress needs to act. That is, the Anthropic position cannot actually constrain the government. If Anthropic were solid, governments would just get unconstrained AI from someone else. Only legislation can create constraints that survive changes in AI suppliers and White House occupiers.
Congress already extensively regulates military procurement through current acquisition laws and annual national defense legislation, imposing conditions on weapons systems, intelligence gathering, and contractor conduct. On the buyer side, Congress could specify which AI systems the military can buy and under what terms. On the seller side, it could dictate what companies would be required or prohibited from incorporating AI systems sold to governments. And it could impose additional transparency and reporting requirements that would give the public visibility into how military AI is actually being used.
The conflict between Anthropic and the Pentagon will be resolved one way or another, either by Anthropic relaxing its terms, by the government finding a replacement, or by some awkward compromise. But without action from Congress, the fundamental problems will remain. The rules governing military AI will be set through ad-hoc negotiations between government officials and individual companies, with no democratic input, no permanent constraints, and no framework that will survive the next administration. That’s not how a democracy should make decisions.
