Seven days ago I was on vacation, watching the sunset on top of a giant boulder, when I got a WhatsApp message from a friend demanding an emergency Friday night podcast about the Trump administration’s standoff with Anthropic. Absolutely not, I thought, and continued to happily ignore the story for another four hours. I had dinner with family, put my kids to bed, and watched a few minutes of the NBA. It was nice. Later that night, though, I went back to my phone to see WhatsApp chats and especially Twitter flooded with reactions to the news.
Much of that Anthropic conversation has continued this week, and I’m obviously back from vacation, but I have to say, I still don’t see any emergency. This post, then, will be more about the reactions, which I have found to be pretty alarmist and hysterical, and above all, ahistorical.
I’m thinking here of China-centric analysts who spent Friday night comparing the Trump administration to the Ming Dynasty burning its fleet of treasure ships, arguing that Anthropic was being treated worse than Alibaba CEO Jack Ma in what’s now the People’s Republic of the United States, and drawing facile connections to chip policy to argue that the U.S. treats Chinese AI companies better than its own. Or take Dean Ball, an architect of the Trump Administration’s AI Action Plan who has since joined the Foundation for American Innovation. He’s been one of my favorite writers on AI, but writing about the Anthropic mess on Monday, Ball lamented: “It is increasingly difficult to honestly discuss the developments of frontier AI, and what kind of futures we should aim to build, without acknowledging our place at the deathbed of the republic as we know it.”
I don’t agree that the republic as we know it is on its deathbed. I will, however, grant that in the modern information environment that claim is more popular than ever. It’s also worth rebutting, which is why I feel compelled to weigh in at all (thoughts on China and the Iran war will come next week).
With respect to Anthropic specifically, Ben Thompson analyzed the situation on Monday and again on Tuesday, and he’s basically landed where I am on all this. I won’t parse too many of the particulars of the situation, but as someone who’s more fluent in law and politics than tech, I’d like to make two specific points about before returning to a more general takeaway.
Anthropic and U.S. History
Melodramatic reactions aside, the stakes of Anthropic’s standoff with the government are indeed dramatic; it’s a big deal to have arguably America’s most successful AI company publicly feuding with the Department of War. Anthropic’s technology is by all accounts incredible, it’s gaining ground throughout corporate America and the consumer space, and Pentagon sources have said, “The other model companies are just behind” when it comes to specialized government applications. In Iran right now, Anthropic’s Claude is being paired with Project Maven to suggest hundreds of targets, issue precise coordinates, and prioritize missions—streamlining weeks of planning in real-time, allowing for faster, more effective warfighting.
On the other hand, if Claude is becoming critical to national security, it really shouldn’t be surprising that the government is not thrilled with the idea of sharing any aspect of its decision-making authority with a group of un-elected researchers with distinct ideas about foreign and domestic policy. And, while the most radioactive aspect of the government’s response is the move to designate Anthropic a Supply Chain Risk, that, too, is not terribly surprising or controversial to me. Yes, the designation at issue traditionally concerns foreign adversary companies deemed a national security threat and has never been used on a U.S. firm, so in that respect it’s technically unprecedented. It would not only eliminate Anthropic’s business with the military, but all military contractors would be prevented from using Anthropic in the course of fulfilling DoW contracts, and possibly all of their business if Hegseth gets his way—though it should be noted that contrary to Hegseth’s tweet announcing this retaliation, the DoW has no statutory authority to interfere with commercial contracts unrelated to DoW work.
Regardless of the mechanics, though, the SCR designation is but one of several powerful levers available to the government, and its use is a signal to everyone that the government intends to use every tool at its disposal to ensure it gets what it wants. Frontier labs will not be nationalized, but they will nevertheless be domesticated and answerable to elected officials. Ball, in his thoughtful post on the matter, correctly decodes the message—a threat of “corporate murder”—and worries that the government’s aggressive posture, “strikes at a core principle of the American republic, one that has traditionally been especially dear to conservatives: private property.”
Yes and no. History is helpful context here. Aside from the irony that “private property” is itself of a legal construct enforced by the same government that according to Ball now imperils the rights of “every investor and corporation in America,” protections of private property in the United States have never been absolute. The entire body of antitrust law, for example, is rooted in laws that contemplate invading corporate property interests for the sake of maintaining healthy market competition and a dynamic economy. And with respect to Anthropic’s interests in relation to the government, there is of course a long history of private firms that have furthered U.S. defense objectives and been rewarded with favorable treatment from the government. AT&T and Bell Labs helped win World War 2 by cooperating on the FIDO torpedo, cooperated with the U.S. on Cold War missile defense systems, helped develop Radar and Sonar systems, and, more controversially, played a prominent role in the NSA surveillance program exposed by Edward Snowden in 2015.
Now, was any of this undertaken under the threat of “corporate murder”? It’s hard to say, and I like to think patriotic sacrifice and deference to the government came more naturally through most of the 20th Century, when existential threats to the American way of life were common. Still, if at any point Bell Labs and later AT&T had flatly refused to accede to government demands, or hinted at restrictions on how their infrastructure could be deployed by the military or its intelligence services, it’s not a great leap to think there could have been consequences for the company’s core business interests and the maintenance of one of history’s greatest monopolies.
Elsewhere, during World War 1, Dupont was allowed to keep its monopoly on government contracts as it powered the war efforts and wound up producing 40% of allied ammunition (the company was later accused of war profiteering). During the Cold War, IBM played a critical role in developing weapons defense systems. During World War 2, companies like Ford and GM were ordered to stop producing cars entirely, producing planes, engines and tanks instead. At the same time, one company supplying uniforms, tractors and autoparts—Montgomery Ward of Chicago—refused to comply with mandated labor orders from the government and had its offices raided by FDR’s Department of War, its CEO hauled away by the National Guard, and all seven of its plants seized by the government. That standoff was adjudicated in a case that Montgomery Ward won. The government appealed, and the matter was ultimately dismissed as moot by the Supreme Court, because then-President Truman returned the property.
Five years later, at the beginning of the Korean War, Truman went the other direction and chose to avert a strike by the United Steelworkers of America by seizing all the major American steel companies and operating those businesses federally so as not to disrupt defense production. That sequence yielded one of the most famous cases in Supreme Court history, Youngstown Sheet & Tube Co. v. Sawyer, a key abrogation of executive authority which held in 1952 that the President lacked the power to seize the mills without express authorization from Congress, forcing Truman to relinquish control of the mills.
70 years after Youngstown, again under claimed exigent circumstances, the Biden Administration was pressuring social media platforms to censor content related to COVID-19 and the vaccine. Those efforts, too, were challenged in court to varying effect, ultimately yielding a Supreme Court decision that found the Plaintiffs lacked Article III standing and never reached the merits of the First Amendment dispute.
Those notes are neither an endorsement nor a condemnation of past government partnerships and/or coercion efforts; the reminder here is merely that a longstanding source of the U.S. government’s strength has been its ability to lean on America’s most successful businesses to do its bidding. There are a variety of checks on abuses in this area (courts, voters, Congress), but every President from Trump to Biden to Truman to FDR has pushed the boundary of executive authority to assert dominance over the private sphere (and you could find 40 more examples). Nothing last week was particularly new, and at no point in the past 100 years has the government’s exercise of its own authority and leverage transformed the nation into a dystopian surveillance state and/or The People’s Republic of the United States. That’s the first point to keep in mind as you read about Anthropic and the U.S. government.
The second point: Consider American history, and then consider that the government is now being presented with a technology that Anthropic CEO Dario Amodei himself claims could “go rogue and overpower humanity,” lead to a spate of “bioterrorism,” or “could devise ways to detect and strike nuclear submarines, conduct influence operations against the operators of nuclear weapons infrastructure, or use AI’s cyber capabilities to launch a cyberattack against satellites used to detect nuclear launches.” American companies do, as a rule, enjoy the freedom to contract and the freedom to refuse to do so; however, any company actively developing that technology is either going to do so in close coordination with the U.S. government, or will risk regulatory reprisal that makes its business and research impossible.
All of that is context for Amodei and Anthropic—the first frontier AI lab to be granted access to classified national security data—responding to the government’s demands not only by refusing to compromise, but by releasing a blog post undermining and likely mischaracterizing the government’s perspective after a private meeting in Washington D.C. Of course there would be a maximalist response to that behavior. My outsider’s guess is that the Supply Chain Risk designation (which became official on Thursday) is less about any contract, and more about the precedent at issue for the U.S. government and all of America’s frontier AI Labs, particularly given Anthropic’s perceived intransigence behind closed doors and extremely public defiance.
I would also guess that Anthropic will ultimately broker a deal with the government. As Charlie Munger said, show me the incentives and I will tell you the outcome. The company is considering an IPO this year, and any U.S. company making weapons-grade technology that maintains an adversarial posture with the American military is going to encounter a variety of material downsides, regardless of Hegseth’s legal authority. “We believe in America,” Amodei said this week, “We believe in helping to defend our country… [We] will continue to work to figure out a solution with DoW.”
Confirmation Bias and the End of the World
Earlier this year I wrote that “if you start from the premise that Trump is simply a boorish moron who’s unfit to hold power in any context, there’s a risk that confirmation bias will distort the analysis of whether any of his policies are reasonable or effective.” Related to that risk: There are millions of Americans who have been conditioned to read the news starting from the premise that Trump and his policies pose a unique, grave and fascist threat to the American system. With that foundation, it’s possible for every new exercise of longstanding government authority to look like distressing evidence that an authoritarian state is upon us
What I find most exhausting about these claims is not the lack of accuracy, but the lack of accountability. The war in Iran will be the end of the American empire until next month, when something else will signal the end. The Supreme Court is deemed an organ of a corrupt fascist government, but no one making those claims adjusts their priors after a 6-3 abrogation of Trump’s tariff authority under IEEPA. The tariffs, themselves, are deemed an act of self-sabotage that will decimate the American economy, but then they don’t, and European governments use them to model tariffs of their own. Paramount under the Ellisons is denounced as an authoritarian propaganda organ after firing Stephen Colbert to curry favor with Trump, but it’s rarely mentioned that Paramount also pays Jon Stewart, and re-signed South Park to call Trump the antichrist every week. We will do this again with Anthropic, of course. Lots of folks have spent the past week warning that the Administration is treating Chinese companies better than American companies and that its tactics will kill investment in American AI; they will forget these conversations happened nine months from now, when America’s lead in AI is even larger.
It would be convenient if this were a problem confined to the Trump era, but it isn’t. Under the Biden Administration we all had to endure years of venture capitalists on Twitter reacting to Biden policies (targeting Elon Musk, censoring social media companies, enforcing 100 year-old antitrust laws) by decrying the onset of authoritarianism and the death of free speech, and/or constantly predicting World War 3. Even under Obama and George W. Bush, their political opponents branded them as not just wrong, but existential threats to the republic. My great hope for the political future is not necessarily any structural reform, but that at some point this century everyone can please just take a breath.
Mind you, my argument here is not that we all should all head to a giant boulder, watch the sunset, and stop paying attention altogether (although let me tell you, it’s great). No: Bad, counter-productive or even outright dangerous policies should be met with full-throated criticism. Trump’s behavior after the 2020 election was a national embarrassment, and World War 3 aside, Anthony Blinken and Jake Sullivan’s foreign policy instincts were genuinely horrific. Today, I think Trump’s foreign policy instincts are much better, but even confined to foreign policy, I despise the Administration’s undignified and often extra-legal, fact-indifferent messaging.
More broadly, and to the point I think Dean Ball was making, I’m not thrilled with the 25 year run of executive branch power consolidation while constitutional hardball grinds Congress to a halt. That trend guarantees that every new Presidential Administration begins with a politicized policy whiplash and an avalanche of executive orders that drastically remake important rules and regulations, and then those EOs work their way through a variety of legal challenges that may or may not alter things further. It sucks!
On the other hand, none of this is fatal. And what drives me crazy is not that news consumers like my mom might believe that we are perpetually on the brink of an unprecedented crisis, but that news purveyors seem to be more invested in making that argument than just about anyone, totally indifferent to the countervailing evidence that undermines their claims. To be clear, then: The United States enjoys tremendous structural advantages with a resource-rich geography, two oceans separating it from any adversaries, and a consumer market, financial system and legal system that each remains the envy of the world. Of all its advantages, the U.S. margin for error might be its most impressive. Speaking of which, the benefit of not being an authoritarian state is that voters provide accountability to everyone in charge and the American system can self-correct as we go. The shape and role of government expanded dramatically under FDR and we all survived; Ronald Reagan arrived and made it smaller, and we survived that, too. There was actually a quite a bit of thriving under each approach.
Catastrophizing current events and predicting imminent American doom are good engagement strategies, but they were a failed electoral strategy for Democrats in 2024, and for anyone trying to understand the world and how it works, they are more likely to cloud people’s judgment than clarify anyone’s thinking. They are also, as a matter of fact and historical record in this country, almost always wrong.
Sharp Text is extension of the Stratechery Plus podcasts Sharp Tech, Greatest of All Talk, and Sharp China. We’ll publish once a week, on Fridays. To subscribe and receive weekly posts via email, click here.