Who Holds the Switch
An AI model did a week of skilled work in half an hour. Three days later one letter switched it off worldwide. The lesson was never the letter.
A few days ago I handed one of the newest and most advanced AI models a job that would have taken a skilled team most of a week. It spun up 124 separate sub-agents, each working its own slice of the problem at the same time, and finished in 32 minutes. By my own rough estimate, something close to sixty hours of trained professional work, done before lunch.
The 124 and the 32 minutes are exact. The sixty hours is my characterization, my way of describing the size of what I watched. Be skeptical of the round number if you want. It will not change the shape of this.
What I saw was not a chatbot writing emails. It was a tool that took a real project of mine, broke it into more than a hundred tasks running at once, and did the work of a roomful of specialists while I refilled my coffee. And it is the slowest, weakest version of this tool I will ever use. Every model after it is faster.
These tools are extraordinary, and I use them every day. I want to tell you about the week one of them vanished, and what that disappearance revealed.
Three days later it was gone
The model in that run was called Fable 5. Anthropic released it on June 9. Three days later, on June 12, it was switched off. Not by the company. By the United States government.
The Commerce Secretary, Howard Lutnick, sent Anthropic a letter invoking national-security powers and placing Fable 5 and a more powerful sibling model, Mythos 5, under export controls (Axios). Because the company cannot check the nationality of every user in real time, the only way to comply was to shut both models down for everyone, everywhere, at once (Anthropic). One letter, and the most advanced version of the most consequential technology of our moment went dark across the world overnight.
Switching to another model took a few minutes, the shrug you give when the plane’s WiFi drops at 35,000 feet. The inconvenience was trivial. What it exposed was not: for one moment the curtain slipped, and you could see how this actually works.
Why this should matter to you
Not because of the letter. Because the thing in that 32-minute run is going to reshape your work, your kids’ job prospects, and the economy you will retire into, faster than any technology before it. And the decisions about how fast it arrives, how safely, who it benefits, and what it is even allowed to do are being made right now by maybe a few dozen people, in rooms you will never enter and fights you will never be asked to settle. You did not vote for them. You will live in the world they build.
That is why the boring-sounding words, export controls and regulation and safety, are not boring. They are the steering wheel of the thing that is about to touch your life whether you follow any of this or not.
The part that is your paycheck
The same power that amazed me is the power that could hollow out the bottom rung of a lot of careers. The man who runs Anthropic, Dario Amodei, has said it about as plainly as a CEO can: AI could wipe out half of all entry-level white-collar jobs in the next one to five years and push unemployment to ten or twenty percent (Axios). He has described the end state as very high economic growth alongside very high unemployment (Bloomberg): the economy booms, and a lot of ordinary people never get into it. This is not one executive musing out loud. This month his company published a formal policy framework that plans for three escalating tiers of disruption, the worst of them severe enough that it floats AI sovereign wealth funds and a common income floor as responses (Anthropic). The people building the thing are writing contingency plans for mass displacement.
Then there are serious people who think that is nonsense. On the All-In podcast this week, the investor David Friedberg called the job-loss story “a Luddite idea that is being disproven every single day,” pointed at strong hiring numbers, and said flatly that “this is a boom, not a bust” (All-In). He runs companies; he says they cannot hire people fast enough.
Nobody can settle this from a podcast, and the early data complicates both stories at once. A Stanford team tracking payroll records for millions of workers found the market is not shedding senior people; it is quietly thinning the bottom rung, with employment for twenty-two to twenty-five-year-olds in the most AI-exposed jobs falling while older and less-exposed workers held steady (Stanford Digital Economy Lab). Friedberg can be right that his firms cannot hire fast enough and the new graduate can still be locked out of the first job that used to train her. The forecasters are guessing, and the ones guessing have billions of dollars riding on how fast this gets deployed. That is the whole reason “roll it out without cratering workers” cannot be left to the people selling it. A laid-off worker is not made whole by a check that arrives after the career is already gone (The Check That Doesn’t Build).
The fight you don’t normally see
This week handed you a rare look at the fight over who controls all of this. Inside the White House, one camp argues that the big AI companies are using safety scare-talk to get rules written that lock in the giants and freeze out everyone smaller, including the free, “open” models anyone can download and run (Sacks). The other camp says the danger is real and the rules are overdue. You do not have to referee that.
The men running these companies are not neutral narrators of the danger. They are businessmen, and excellent ones. Dario Amodei and Sam Altman are not villains (hopefully) and they are not fools. They are doing exactly what great businessmen do, which is champion their companies. The mistake is to hear them and assume they are championing you.
A frontier AI lab’s interest is genuinely served by two things: rules that raise the cost of competing, and open models never catching up. So when the men who run those labs call for the government to step in, it might be wise and it might be self-serving, and from where you sit those two things look identical. You do not have to assume bad faith. You just have to remember whose job it is to win. It is the same lesson Wall Street teaches eventually, that the analyst talking up a stock is Pushing Their Book: talking his own position, which is normal, and the only error is believing the book is about you.
The question underneath it is the one in You’ll Own Nothing: who gets to hold the most important thing of the century while it is still being built, and on what terms the rest of us are allowed near it.
Why it got switched off
So why did the government pull the switch? More of the story has come out since, and it is messier than either side first let on.
The stated reason was a security scare. A tester got the more powerful model to hand over information that could aid a cyberattack, and when the administration asked Anthropic to fix it or pause the release, the company declined and shipped anyway (Axios). By later reporting, the alarm came from Amazon, Anthropic’s own largest backer: CEO Andy Jassy flagged the flaw to the White House, which is how a private safety dispute became a federal one (Fortune). Anthropic says the flaw was narrow, not universal, and no worse than what already exists in rival models (Anthropic). Critics in the press read it differently: the government has been feuding with this company for months, and to them the shutdown looks like punishment wearing a security badge (Fortune, NYT). Others say it was simply inevitable, that frontier AI is now treated like nuclear material and was always going to get locked down.
Whatever the reason, and it may be all of them at once, one fact survives every version of the story: a single official switched off a frontier tool for the entire world, overnight, with no public process and, by some reporting, not even a written explanation of what the danger was (BankInfoSecurity). And it has stayed off. Anthropic flew its leaders to Washington; the talks ended with no agreement and no shared definition of what fixing it would even require, and as I write this the models are still dark (IAPP). A power that large should not depend on us trusting that it happened to be used wisely this one time.
Where I land
I want the United States to win this—and I don’t exactly trust the people most likely to win it. I want these companies to keep breaking frontiers, and handing the lead to anyone else would be worse. The irony of the shutdown is hard to miss: in the same days we switched off our own frontier models for everyone outside the country, a freely downloadable Chinese model climbed to the top of the major benchmarks, matching or beating the American leaders at a fraction of the cost (VentureBeat). You do not protect a lead by unplugging it. And leading the world in a technology this powerful is hollow if it wrecks the workers it was supposed to lift, or if the handful of people steering it answer only to themselves. We are allowed to want both: the edge, and a rollout built for ordinary people instead of merely past them.
You do not have to lose your mind because a model went dark for a week. I did not. You do have to refuse to look away, and to expect, from the companies and the government alike, that they build this for you and not simply past you. Very few hands are on the switch. The least the rest of us can do is keep watching the hands.
There is a deeper question hiding under all of it, the one the builders are quietly answering whether they admit it or not: what a human being is actually for. That one I will take up on Saturday.
Sources
Axios, “Trump admin blocks foreign access to Anthropic’s most advanced AI models,” 2026-06-12, https://www.axios.com/2026/06/12/anthropic-trump-mythos-fable-national-security
Anthropic, “Access to Fable and Mythos” (official statement), https://www.anthropic.com/news/fable-mythos-access
Fortune, “A warning from Amazon led the White House to shut down Anthropic’s Mythos model” (Andy Jassy flagged the Fable 5 jailbreak to the administration), 2026-06-14, https://fortune.com/2026/06/14/how-a-warning-from-amazon-led-the-white-house-to-shut-down-anthropics-mythos-model/
Fortune, “Anthropic disables Fable and Mythos after export controls cite national security threat,” 2026-06-13, https://fortune.com/2026/06/13/anthropic-disables-fable-mythos-export-controls-national-security-threat/
The New York Times, “Trump Administration Restricts Anthropic’s Most Advanced AI Models,” 2026-06-13, https://www.nytimes.com/2026/06/13/us/politics/trump-anthropic-ai-models.html
The Wall Street Journal, “Anthropic Halts Access to Top AI Models After U.S. Ban on Foreign Use,” https://www.wsj.com/tech/ai/anthropic-halts-access-to-top-ai-models-after-u-s-ban-on-foreign-use-a4bca2cc
BankInfoSecurity, “US Pulls the Plug on Anthropic’s Top AI Models,” https://www.bankinfosecurity.com/us-pulls-plug-on-anthropics-top-ai-models-a-31964
IAPP, “The global implications of the White House’s export controls on Anthropic” (talks ended without resolution; models remain suspended), https://iapp.org/news/a/the-global-implications-of-the-white-houses-export-controls-on-anthropic
Axios, “Behind the Curtain: Top AI CEO foresees white-collar bloodbath” (Amodei: half of entry-level white-collar jobs at risk in 1–5 years; 10–20% unemployment), 2025-05-28, https://www.axios.com/2025/05/28/ai-jobs-white-collar-unemployment-anthropic
Bloomberg, “Inside Anthropic, the $965 Billion AI Juggernaut | The Circuit” (Amodei on very high GDP growth alongside very high unemployment),
Anthropic, “Economic Policy Framework” (three-tier response to AI labor disruption, up to sovereign wealth funds and a common income floor), 2026-06, https://www.anthropic.com/policy-on-the-ai-exponential/epf
Stanford Digital Economy Lab, “AI and Labor Markets: What We Know and Don’t Know” (entry-level employment in AI-exposed jobs falling for workers aged 22–25 while older and less-exposed workers hold steady), https://digitaleconomy.stanford.edu/news/ai-and-labor-markets-what-we-know-and-dont-know/
All-In Podcast, “Anthropic’s Fable Backlash, Nationalizing AI, Inflation Heats Up & California’s Broken Elections,” 2026-06-13 (David Friedberg: “a Luddite idea that is being disproven every single day,” “this is a boom, not a bust”),
David Sacks (@DavidSacks), posts characterizing Anthropic as “woke” / “doomers” / regulatory capture,
VentureBeat, “Z.ai’s open-weights GLM-5.2 beats GPT-5.5 on multiple long-horizon coding benchmarks for 1/6th the cost,” https://venturebeat.com/technology/z-ais-open-weights-glm-5-2-beats-gpt-5-5-on-multiple-long-horizon-coding-benchmarks-for-1-6th-the-cost
Miles DeBenedictis, The Check That Doesn’t Build, https://pastormiles.substack.com/p/the-check-that-doesnt-build
Miles DeBenedictis, Pushing Their Book, https://pastormiles.substack.com/p/pushing-their-book
Miles DeBenedictis, You’ll Own Nothing, https://pastormiles.substack.com/p/youll-own-nothing
This article was developed using AI writing tools I built to work with my voice, research, and editorial framework. The ideas, arguments, and positions are mine. The pipeline that helps me draft, evaluate, and refine them is something I created as part of my work at Nomion AI. I believe in building with AI and being honest about it. If you want to know more about that process, ask me.
“Very few hands are on the switch. The least the rest of us can do is keep watching the hands.”
Amen. I can’t wait till Saturday.