Picture the most powerful AI an American company has ever released to the public. Out for three days. Three. Then the federal government, in a fit of what it calls national security, orders it yanked offline so that no foreign national can touch it. The order is written so sloppily that to comply, the company has to shut the thing down for *everyone* — including the American citizens the order was supposedly protecting.
That's not a hypothetical. That happened last week. And it is one of the dumbest things I have watched this government do, which, given the competition, is a hell of a sentence to have to write.
Let's walk through the nutfuckery, because every single layer of it is worse than the last.
What actually happened
On June 9, Anthropic released Fable 5 and Mythos 5, the most capable models it had ever put in public hands, a tier it calls "Mythos-class," sitting above everything that came before. Three days later, on June 12 at 5:21 in the evening, the Commerce Department sent a letter. The U.S. government had issued an export control directive ordering Anthropic to suspend all access to Fable 5 and Mythos 5 by "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees."
Read that scope again, slowly, because the stupidity lives in the details. Not foreign governments. Not adversaries. Not China. *Any foreign national, anywhere on Earth, including Anthropic's own employees who happen not to hold a U.S. passport.* The British. The Canadians. The French. Your tech buddy in Toronto. The non-citizen engineer who helped build the damn thing.
The directive was so absurdly broad that Anthropic said it had no way to comply except to pull the plug entirely. So they did. They disabled the models for everybody. Which means the practical result of an order justified as protecting American national security was that Americans lost access to the most powerful American AI, by order of their own government. They set out to lock the foreigners out of the house and managed to brick the front door for the people who live there.
If you tried to write a satire of government incompetence, an editor would send it back for being too on the nose.
The barn door was already wide open
Here's the part that elevates this from "clumsy" to "genuinely does not understand the thing it is regulating."
You cannot un-release a software model the way you recall a defective toaster. The government's stated worry was a "jailbreak" — a method of bypassing Fable 5's safety guardrails. Set aside for a second whether that worry holds up. Even if it were the crisis they claim, the model had already been out for three days to a global user base. Anyone motivated to grab the weights or document the technique had a 72-hour head start. Recalling it afterward is closing the barn door, lighting the barn on fire, and then issuing a press release about how seriously you take barn security while the horse is three counties away.
And it gets stupider, because the jailbreak panic itself doesn't survive contact with the facts. Anthropic looked at the technique the government was waving around and said it amounted to a small number of previously known, minor vulnerabilities — essentially asking the model to read a codebase and point out software flaws. The company's position: this is a narrow, non-universal issue, the kind every safeguarded model in the industry is vulnerable to, and recalling a commercial product deployed to hundreds of millions of people over it is insane. You don't have to take the company at its word. You can take the outside experts, who weren't much kinder to the government.
And no one else got this treatment
Now here's the load-bearing absurdity, the one that turns "incompetent" into "incompetent *and* selective."
Anthropic pointed out that the exact same jailbreak could likely pull similar capabilities out of other publicly available models — they named OpenAI's GPT-5.5 specifically — and *those* models are not sitting under any national security export control. So the government identified a category of risk, found it in one American company's product, and then slapped that one company while leaving its competitors, who have the same vulnerability, completely untouched.
That's not a security policy. A security policy addresses the risk wherever it lives. This addressed the risk at exactly one company and called it a day. Either the government doesn't understand that the vulnerability is industry-wide, which is incompetence, or it understands perfectly and is using "national security" as a costume for going after one specific company, which is something worse.
Dean Ball, an AI policy expert who briefly worked in the Trump administration and is no reflexive critic, called the whole thing "cartoonish." His point landed like a slap: this is an administration whose posture is that we *should* be exporting advanced AI chips to China, and that same administration turned around and tried to ban Britain — and every other non-American on the planet — from using our best AI model. Sell the hardware to the rival superpower, but God forbid a Canadian touch a chatbot. "I have no words," Ball wrote. Neither, frankly, do I, and you know that's rare for me.
The part where it stops looking like stupidity and starts looking like a grudge
So far this reads as pure incompetence. But there's a backstory, and once you know it, a colder possibility moves into frame.
This administration and Anthropic have been at war since February, when Trump and Defense Secretary Pete Hegseth publicly cut ties with the company after it refused to allow unrestricted military use of its AI. And what did Anthropic refuse, specifically? It declined to let its model be used for lethal autonomous weapons without human oversight, and for mass surveillance of Americans. Pause on that. A company said "we won't build you killer robots that operate without a human in the loop, and we won't help you spy on your own citizens," and the government's response was to brand it a "supply chain risk to national security" and order federal agencies to stop using its product.
Then this export ban lands on the same company. And a federal judge — District Judge Rita Lin, hearing Anthropic's challenge — said from the bench that it "looks like an attempt to cripple Anthropic," and voiced concern that the government might be punishing the company for openly criticizing it.
When a sitting federal judge looks at your national security order and wonders out loud if it's actually retaliation, the security justification has officially left the building. This is starting to look less like a government carefully guarding the nation and more like a powerful man with a bruised ego using the machinery of the state to kick a company that told him no. The incompetence might not even be the worst part. The assholery might be.
I'll hold the honest tension, because that's what we do here
I'm not going to pretend Anthropic is a flawless victim, because that's not the truth and you'd smell it. This is a company that called its own model too dangerous to release widely, that has loudly demanded tighter AI regulation, that built its whole brand around being the cautious one in the room. There's a real argument that they wrote the legal predicate for their own punishment — talk loudly enough about how dangerous your technology is, and don't be shocked when the government takes you at your word and reaches for a hammer.
Fine. Hold that tension. You can think frontier AI genuinely needs guardrails *and* recognize that this particular intervention was a botched, selective, almost certainly vindictive mess. Both are true. The existence of a real problem does not make every government response to it intelligent. Sometimes the house really is on fire and the fire department still shows up, hoses the wrong building, tickets the homeowner, and drives off feeling heroic.
The precedent is the actual danger
Here's why this matters beyond one company's bad week.
This is the first time the U.S. has used export controls on an AI *model* itself, not the chips underneath it. That's a new tool, and we just watched its debut performance: overbroad, technically illiterate, applied to one company while its competitors skated, and credibly suspected of being personal. That's the template now. That's the precedent sitting on the shelf for next time.
Governments reaching in to switch off released technology on a whim, with no apparent grasp of how the technology works, is not strength. It's the opposite. It's isolationism that boomerangs straight into the citizens it claims to defend — Americans denied American tech, foreign allies shoved toward building their own competitors precisely because we made ours unreliable, and the global lead this administration says it wants to protect quietly handed to anyone willing to be less erratic. You don't win a technology race by yanking your best runner off the track three steps in because you got into a slap fight with him in the locker room.
The world is watching how America governs the most powerful tools it builds. Last week it watched us point a national security weapon at a chatbot, miss, and shoot ourselves in the foot. Allies don't see toughness in that. They see a government that doesn't understand its own industry, can't write a coherent order, and might be doing the whole thing out of spite.
That's not a superpower protecting its edge. That's a clown car with a Commerce Department letterhead, and the precedent it just set will outlast the tantrum that produced it.
Not left. Not right. Just tired of liars pretending we're stupid.
Sources
• Anthropic, "Statement on the US government directive to suspend access to Fable 5 and Mythos 5," June 12, 2026 (anthropic.com/news/fable-mythos-access)
• CNBC, "Anthropic disables access to Fable 5 and Mythos 5 to comply with government directive," June 12, 2026 (5:21pm ET timing; Project Glasswing background)
• Fortune, "Anthropic disables Fable and Mythos AI models after U.S. government bars it from giving foreigners access," June 13, 2026 (Commerce Department; scope including non-citizen employees; GPT-5.5 comparison; Dean Ball "cartoonish" quote)
• TIME, "Anthropic Pulls Its Most Powerful AI Models After U.S. Bars Foreign Access," June 13, 2026 (first use of export controls on models rather than chips; "Mythos-class" framing)
• Al Jazeera, "US orders Anthropic to disable AI models for all foreign nationals," June 13, 2026
• Business Standard, "Why US has restricted foreign access to Anthropic's Claude Fable 5, Mythos," June 14, 2026 (jailbreak described as minor, previously known vulnerabilities)
• TechPolicy.Press, "Anthropic's Mythos Recall and the White House's Missing AI Safety Playbook," June 2026 (WSJ reporting on Andy Jassy's role; Amodei policy context)
• The Hill, "Anthropic withdraws AI models Fable, Mythos due to export controls," June 2026 (negotiations; DoD relationship)
• Euronews, "'An attempt to cripple Anthropic': US judge questions whether ban on AI company is punitive," March 25, 2026 (Judge Rita Lin; February ties-cutting over lethal autonomous weapons and mass surveillance refusals)
