On Anthropic, Trump Administration Revives Bill of Attainder–Style Extortion

This is thuggish behavior.

O bviously, there is a lot going on in the world, but even with war now raging in the Middle East, none of what’s happening may be more consequential than the Trump administration’s extortionate attempt to kill a top American technology company — one that has provided essential assistance to national security. The president and the Defense Department are in a snit over a contract dispute with the tech giant Anthropic, involving how controversial artificial intelligence (AI) capabilities are incorporated in our national defense.

The AI at issue is derived from “Claude,” as Anthropic large language models are collectively known. The dispute is a critical one. It illustrates how the rapid evolution of AI is racing beyond our law’s capacity to keep up — to even grasp what’s at stake, much less do sensible regulatory line-drawing.

Established in 2021, Anthropic over a year ago formed partnerships with government contractors — e.g., Palantir Technologies, a principal government software supplier. As a result, Claude was incorporated into the operating systems of classified U.S. intelligence and defense systems. The DoD uses Anthropic’s AI for various purposes, including combat operations and surveillance. Its functions are top of the line, and they have worked well.

Understandably, the DoD wants the discretion to use technology it purchases for any lawful purpose. The law, however, is underdeveloped and doesn’t address — or, as DoD sees things, currently appears to permit without limitation — the armed forces’ operation of weapons systems that are autonomously controlled by AI rather than a human agent, and the government’s incorporation of AI to conduct mass surveillance — which could theoretically include wholesale monitoring of Americans (on the latter, see Jim’s must-read Jolt on Friday).

Needless to say, these are controversial uses. Anthropic, as is its right as a private company, does not want its AI used in these manners. The Trump DoD, at least on the issue of mass surveillance, reportedly assured Anthropic that it would not use the technology in such an objectionable manner but has resisted putting that restriction in writing — which, to say the least, calls DoD’s earnestness into question.

The parties reached an impasse. That’s life in business: Anthropic has good faith concerns that it will be blamed if the government uses its technology abusively; the administration reasonably counters that policy on the AI frontier ought to be made by politically accountable government officials, not Silicon Valley CEOs. On brand, the Trump administration believes policy ought to be decreed unilaterally by the executive branch rather than Congress — or, to give them the benefit of the doubt, Trump officials take congressional silence as a choice, rather than as a function of AI’s dynamism outpacing lawmakers’ sclerosis, and believes no regulation means there must be no restraints. (Of course, even when Congress or the Constitution expressly limits executive power, the Trump administration often construes Article II as giving the president inherent power to ignore them.)

While the DoD and Anthropic were grinding toward their impasse, it appears that OpenAI (which operates ChatGPT) may have swooped in and agreed to give the DoD what it wanted if the government were to supplant Claude with ChatGPT. We’ll have to look at the fine print: OpenAI’s CEO, Sam Altman, appeared to sympathize with the concerns expressed by his competitor, Anthropic CEO Dario Amodei. While Altman then cut a deal with the DoD, it’s not clear that OpenAI fully committed to do what Anthropic had declined to do.

In any event, that’s life in business, too.

On the decision to stop doing business with Anthropic, the DoD is well within its rights. Policy-wise, it may be a bad decision. I’m not in a position to evaluate that because I don’t know enough about how great the value of Claude’s services are to the government, such that I could competently weigh that against what is lost by the two restrictions Anthropic wanted to impose. And I don’t know how much the DoD will be set back, at least for a time, by having to transition complex systems from Claude to ChatGPT. All I can confidently address is the law: If a business does not want to provide an unrestricted service to the government, and the government does not want to pay for that service unless it is unrestricted, neither party is obliged to do business with the other.

That should be the end of the story. But it is not.

That’s because, in the Trump administration’s statist pique over Anthropic’s insistence on the two restrictions, Defense Secretary Pete Hegseth pressured Amodei to capitulate to the government’s demands by the close of business Friday or else. Initially, that ultimatum included a threat to invoke the Defense Production Act (DPA) to, in effect, nationalize Anthropic’s business.

In the end, he didn’t go that far — at least not yet. Once the deadline passed, however, Hegseth announced (on X, of course) that the government is designating Anthropic a “supply chain risk to national security” (SCR). As a result, the president has directed that all federal agencies and federal contractors stop using Anthropic’s AI.

Hegseth publicly accused Anthropic of “arrogance and betrayal” and of “a cowardly act of corporate virtue signaling that places Silicon Valley ideology above American lives.” Amodei is a political progressive who is adherent to “effective altruism” (the preening commodification of corporate charitable giving that was the signature calling of the now-convicted mega-fraudster, Sam Bankman-Fried). Hegseth derided Amodei’s concerns about autonomous battlefield weapons and mass surveillance as “defective altruism,” which he brayed would “never outweigh the safety, the readiness, or the lives of American troops on the battlefield.”

Not to be outdone, President Trump chimed in (on his private media company’s Truth Social platform) about the “Leftwing nut jobs at Anthropic,” warning that he would “NEVER ALLOW A RADICAL LEFT, WOKE COMPANY TO DICTATE HOW OUR GREAT MILITARY FIGHTS AND WINS WARS!” because that “decision belongs to YOUR COMMANDER-IN-CHIEF.” (Note: The president is not “your commander in chief”; the Constitution gives him command of the armed forces; as to the American people, the president is their servant, not their master.)

Trump added:

I am directing EVERY Federal Agency in the United States Government to IMMEDIATELY CEASE all use of Anthropic’s technology. We don’t need it, we don’t want it, and will not do business with them again! There will be a Six Month phase out period for Agencies like the Department of War who are using Anthropic’s products, at various levels. Anthropic better get their act together, and be helpful during this phase out period, or I will use the Full Power of the Presidency to make them comply, with major civil and criminal consequences to follow.

This is thuggish behavior. Note that the DPA, which Hegseth threatened to use, requires a presidential finding that goods or services at issue are necessary for national defense (see §4511); yet, Trump says of Anthropic’s product, “We don’t need it. We don’t want it.”

Moreover, it has been reported that, while Hegseth was first threatening to brand Anthropic an SCR, the Pentagon’s chief technology officer, Emil Michael, was offering it such concessions as written commitments about surveillance and autonomous weapons, as well as inviting it to participate in the deliberations of the DoD’s ethics board.

And remember, this is an administration that — after being given a piece of the action, despite the absence of congressional revenue-raising authorization — has authorized Nvidia to sell some of the government’s microchip technology to China, America’s main geopolitical rival. And, after agents of the United Arab Emirates poured $2.5 billion in investments in and purchases from the Trump family’s private cryptocurrency business, the administration approved chip sales to the UAE — despite prior objections by U.S. national security officials and Congress that its collaboration with China on dual-use technology warranted a prohibition on chip sales to the UAE.

But now the administration would have us believe that Anthropic, which has worked closely with the DoD to promote national security against the threat of China, is a risk to supply chains.

This is ridiculous. Federal law (§3252, of Title 10, U.S. Code) defines SCR as:

[T]he risk that an adversary may sabotage, maliciously introduce unwanted function, or otherwise subvert the design, integrity, manufacturing, production, distribution, installation, operation, or maintenance of a covered system so as to surveil, deny, disrupt, or otherwise degrade the function, use, or operation of such system.

The Trump administration has not claimed that Anthropic poses risks that could enable hostile regimes to compromise American defense systems. To the contrary, Trump and Hegseth have been pressuring Anthropic to continue in what has been close collaboration. This is not about supply chain risk; the administration is angrily lashing out because Anthropic has declined to provide it greater latitude to incorporate its AI into American defense systems.

Furthermore, Section 3252 mandates that, before an SCR designation can be made, the government must (a) make various findings, including that the designation is necessary to protect national security (which the president’s comments show it is not) and that less intrusive measures to reduce risk are unavailable; and (b) notify Congress. Does anyone believe Hegseth and other administration officials have complied with these requirements — or that they could have done so in good faith — before announcing the SCR designation?

This extortionate move is materially indistinguishable from the executive orders by which the president illegally sought to put law firms he didn’t like out of business in the first weeks of the administration. Several judges have struck down these extortionate directives (see, e.g., here and here.) At the time, I described them as, functionally, unconstitutional bills of attainder or bills of pain and penalty: provisions that condemn a person or category of persons for alleged wrongs and impose penalties, all without trial or due process. (The provisions the Framers were referring to were legislative acts; the wrong is the same, nevertheless, whether it is accomplished by an unconstitutional statute or unconstitutional executive action.)

In essence, the Trump administration is trying to destroy the business of Anthropic, a $380 billion company, by leveraging the government’s vast array of procurement authorities and contractual arrangements. On that, it’s worth reading a post yesterday by Dean W. Ball of the Foundation of American Innovation, who advised the administration on AI technology:

Nvidia, Amazon, Google will have to divest from Anthropic if Hegseth gets his way. This is simply attempted corporate murder. I could not possibly recommend investing in American AI to any investor; I could not possibly recommend starting an AI company in the United States.

He’s right. Congress and the courts must not allow this to stand. Again, if the administration no longer wants to work with Anthropic, that’s within the president’s discretion even if he exercises that discretion for all the wrong reasons. But a vindictive executive effort to destroy a company that has chosen not to provide the government with unrestricted use of its products — and for entirely defensible reasons, even if it doesn’t need such reasons in a free society — is intolerable.