That's the Point
A story of what happens when the people enforcing the law are unencumbered by laws.
On Friday, June 12, at 5:21 in the evening, the Commerce Department sent a letter to Anthropic, makers of the chatbot Claude. The letter ordered the company to suspend access to its two most advanced AI models for every foreign national, inside or outside the United States, including Anthropic’s own employees. Anthropic had no way to verify citizenship across its user base, so by 10 PM both models were dark for everyone on earth.
When CEO Dario Amodei spoke to Commerce Secretary Howard Lutnick that evening, he lamented the ban meant they couldn’t serve the models at all. “That’s the point,” Lutnick responded, according to people familiar with the call.
Three weeks earlier, Andrej Karpathy, formerly head of AI at Tesla, joined the team at Anthropic. He can’t touch those two models now. He was born in Slovakia, so his job at the company is impossible to do.
An unelected cabinet secretary shut down a product serving hundreds of millions of people. The justification was a jailbreak: a researcher found a way around the model’s safety guardrails, and the government flinched.
No law authorized the process. No court reviewed the evidence. No appeal process existed for Anthropic to use. Anthropic says the issue was minor, but there’s no neutral body to settle the question.
Donald Trump has the most aggressively anti-regulation administration in modern history when it comes to AI. But it shut down a commercial AI product overnight with a single letter, because it could. Because there’s nothing that says it can’t.
The Preemption
How was this letter possible?
When Trump took office back in January of 2025, he aimed to undo Biden’s legacy. One of the pieces of that legacy was the Biden era’s executive order on AI. That order required the top AI companies to share their safety evaluations with the government before they could release new models.
Before Trump took office, there was a process. After, there was none at all.
With no federal framework in place, states took it upon themselves to draft legislation regulating AI. In the 2025 legislative session, every state in the union introduced some kind of AI bill. The administration treated that like a threat. David Sacks, the administration’s AI czar, cited “1,200” bills and warned about “fifty states going in fifty different directions.” The actual numbers were less dramatic. Only 136 of those bills were enacted, and only 26 directly regulated AI companies. The response was a Justice Department task force and a threat to pull federal funding.
That summer, the White House presented its formal AI strategy: the AI Action Plan. The plan’s premise was simple. Premature regulation could slow American AI development, so the government should hold back and let innovation run.
Later that winter Trump released yet another executive order pressuring states to stop passing AI laws. The order created an AI Litigation Task Force inside the Justice Department, dedicated to challenging state regulations. It directed agencies to review whether states receiving federal grants had passed AI laws that conflicted with the administration’s framework, and to consider pulling funding if they had. Colorado had passed a law banning algorithmic discrimination. California had passed a transparency law requiring safety disclosures for frontier models. The executive order put both in the crosshairs.
The Center for AI Standards and Innovation (CAISI), the federal government’s one body publicly evaluating AI models, had its director pushed out and was ordered to stop publishing its findings. The public’s only independent window into how these models actually perform went dark.
Each move was presented as deregulation. Each one removed something specific: a reporting requirement, a transparency standard, a state consumer protection, a public evaluation.
The Turn
A surface-level read of this situation sounds comforting. The administration believes in free markets. Maybe they went too far on one bad Friday. They’ll correct course. They still want deregulation.
That reading only works if you ignore the pattern.
Every AI law that could have been written would have come with constraints attached. Defined criteria for enforcement. Due process requiring the government to show its reasoning. Judicial review. Appeal mechanisms. Transparency requirements. All of that constrains the executive, because a government operating under a law has to follow it. The rules bind the regulator as much as the regulated.
By clearing every one of those constraints out of the space, the administration freed itself.
“We don’t regulate AI” sounds like the government stepping back. What it means in practice is that when the government steps forward, there’s nothing in its way.
The Squeeze
The export control didn’t come out of nowhere. Anthropic has been in a fight with the federal government since January, and the fight has always been about the same thing.
Earlier this year Anthropic had a contract with the Pentagon worth up to $200 million. The Department of War set the terms: Claude could be used for all lawful purposes.
Anthropic balked. They drew two red lines: Claude could not be used for domestic surveillance of US citizens and it could not be used for fully autonomous weapons.
War Secretary Pete Hegseth responded with a public ultimatum: agree to unrestricted access by 5:01 PM Friday or he’d use the Defense Production Act to force them to comply.
Amodei refused. Publicly. “We cannot in good conscience accede to their demands.”
Trump responded on Truth Social: “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.”
Then the consequences started arriving.
The Pentagon designated Anthropic a supply chain risk, a label reserved for foreign adversaries and compromised foreign vendors. It had never been applied to an American company. A federal judge called the designation retaliatory, the reasoning “Orwellian.” The appeals court reversed her injunction, and the designation held.
Hours after the ban, OpenAI announced a Pentagon deal. Caitlin Kalinowski, OpenAI’s lead for robotics and consumer hardware, resigned, saying surveillance and autonomous weapons “are lines that deserved more deliberation than they got.” At an all-hands meeting, Altman told his staff they don’t “get to make operational decisions” about how the military uses their technology.
The Pentagon signed AI deals with eight companies. Anthropic’s enterprise customers started getting nervous. Defense contractors dropped Anthropic’s products.
Meanwhile, Anthropic was preparing to go public. On June 1, the company filed its IPO paperwork with the SEC after a $65 billion funding round that valued it at $965 billion. Eight days later, it launched Fable 5 to demonstrate its capabilities to the public markets. Eleven days after filing, the government shut both models down. Every investor evaluating the company just watched the US government demonstrate it can kill the flagship product with a single letter.
When Anthropic protested the export control on their product, showing the jailbreak was narrow and minor, the government moved the goalposts. Fix all jailbreaks, the White House told them, across all frontier models, and flag every one to the government yourself.
Independent cybersecurity experts say what the administration is demanding can’t be done. Perfect jailbreak resistance is impossible. The government knows this. Their own evaluators know this. Anthropic said so in the model’s own launch materials.
A standard that’s impossible to meet becomes a lock. The model stays dark until the government decides it doesn’t, and there’s no law defining what “enough” looks like. The original demand from February, unrestricted military access, has never left the table.
The only company that drew red lines is the one being squeezed until it erases them.
Behind The Lines
Here’s the part that makes the whole picture legible.
While all of this was playing out in public, six Anthropic engineers were sitting inside the National Security Agency. They’ve been there since April, deploying Mythos, Anthropic’s most powerful model, for offensive cyber operations on classified networks. The Financial Times broke the story in early June. The NSA is using Mythos to find exploitable vulnerabilities in software, including Microsoft products, and according to one source, to prepare for infiltrating networks in countries like China and Iran.
The Pentagon designated Anthropic a supply chain risk in March. The NSA started deploying Anthropic’s model weeks later. Same parent agency. Pentagon CTO Emil Michael explained the apparent contradiction to CNBC by carving the product away from the company: the supply chain risk designation still stands, he said, and Mythos is a “separate national security moment.”
They banned the company. They kept the product.
The model that’s too dangerous for the public, too dangerous for foreign nationals, too dangerous for Karpathy to touch at his own company, is running fine inside the intelligence community’s classified networks.
And the Department of War has been using Anthropic’s models to support military operations in the Iran war, while simultaneously arguing in court that using Anthropic’s tools threatens national security.
The Ledger
A vacuum with no rules is useful to whoever fills it. So it’s worth asking who benefits.
Trump’s Office of Government Ethics filings for the first quarter of 2026 show 3,642 individual stock trades. The portfolio is heavily concentrated in AI companies: Nvidia, Palantir, Dell, Oracle, Microsoft, AMD, Amazon, Apple, Broadcom. Eric Trump has called the assets a “blind trust” invested in “broad market indexes.” Thirty-six hundred trades in individual stocks is not a broad market index.
Some of the trades line up with government actions in ways that are hard to ignore. Trump bought Palantir stock across three months, then praised the company by ticker symbol on Truth Social. He bought Dell, then praised Dell at a White House event. He bought Nvidia shortly before the company expanded an AI deal that required regulatory approval. A report from the watchdog group Public Citizen, published June 4, found that 16 of the 27 corporate donors to Trump’s White House ballroom project are facing federal enforcement actions that have been suspended by the administration, including antitrust cases against Amazon, Apple, Meta, Microsoft, and Nvidia. Fourteen of those 27 received new or expanded government contracts totaling over $50 billion.
Anthropic is a private company preparing for its IPO. Trump can’t hold the stock. He holds stock in every publicly traded competitor.
Follow that through the timeline. In February, the Pentagon banned Anthropic from all federal contracts. OpenAI had a replacement deal within hours. In May, the Pentagon signed AI agreements with eight companies: OpenAI, Nvidia, Oracle, Microsoft, Amazon, SpaceX, Google, and Reflection. Anthropic was excluded. In June, the export control shut down Anthropic’s flagship product for every customer on earth. Every enterprise client shopping for a frontier AI model that week was shopping Anthropic’s competitors.
Every action the administration has taken against Anthropic has moved market share toward companies in the president’s portfolio. I’m laying out the transactions. You can do the math.
The Inheritance
I’ve spent a year arguing that existing law was sufficient to govern AI in most cases. I still believe that. Very few new laws need to be written. The legal tools to handle fraud, discrimination, product liability, surveillance, and civil rights already exist. The White House’s position aligned with my own.
The difference is what “sufficient” means. I meant the existing law should be applied. What they meant was that law should stay absent, so nothing can constrain their discretion. Same bumper sticker. Different meaning.
Dean Ball, the person who drafted the administration’s AI Action Plan, now calls its execution “the antithesis of the rule of law.” He watched the framework he built get used as a weapon and said so publicly because the alternative is complicity.
The vacuum has always been occupied. The evidence is in the stock trades, the threats, the classified networks, and the contracts that followed every punishment.
Every future administration inherits this vacuum. Every power this White House has exercised over AI, the export controls, the supply chain designations, the ability to shut down a product with a Friday letter, exists because there’s no law that says it can’t. There’s no law because this administration made sure there wouldn’t be one.
They took Fable away. They can do it again. And so can whoever comes next.
Enjoyed this piece?
I do all this writing for free. If you found it helpful, thought-provoking, or just want to toss a coin to your internet philosopher, consider clicking the button below and donating $1 to support my work.
“William Roper: “So, now you give the Devil the benefit of law!”
Sir Thomas More: “Yes! What would you do? Cut a great road through the law to get after the Devil?”
William Roper: “Yes, I'd cut down every law in England to do that!”
Sir Thomas More: “Oh? And when the last law was down, and the Devil turned 'round on you, where would you hide, Roper, the laws all being flat? This country is planted thick with laws, from coast to coast, Man's laws, not God's! And if you cut them down, and you're just the man to do it, do you really think you could stand upright in the winds that would blow then? Yes, I'd give the Devil benefit of law, for my own safety's sake!”
― Robert Bolt, A Man for All Seasons: A Play in Two Acts
Meaning. They’re coming for that next. Edit: collectively.