Make no mistake – this is not a gadget or novelty computer tech: The most powerful artificial intelligence ever available to the general public is built on a system smart enough to find security holes in every major operating system on Earth. 72-hours after it launched, the federal government ordered it shut off. Due to the nature of this particular cutting-edge “AI software” it’s important to know why.
On Friday evening, June 12, at 5:21 p.m. Eastern, a directive from the U.S. Commerce Department arrived at the offices of the artificial intelligence company Anthropic, developers of ‘CLAUDE AI’. By the end of the night, the most capable AI system ever offered to the general public had gone dark. It did not crash. The company did not retire it. The United States government ordered it switched off, and Anthropic complied.
The system is called Claude Fable 5. Three days earlier, it had launched with the kind of muted reception that now defines the AI era: a few headlines, a few impressed engineers, and a general public that mostly shrugged. For most people, a new AI model lands somewhere between a new smartwatch and a VR headset on the excitement scale. A curiosity. A novelty. Something other people seem worked up about. Meanwhile, quietly, hundreds of millions of people have folded these tools into their daily work and lives. It is one of the largest technological shifts in a generation, and much of the public still files it under “gadget.”
That gap, between what these systems can actually do and how casually the world treats them, is the whole story here. Because the government does not treat Claude Fable 5 like a gadget.
The Most Capable Claude Ever Built
By its maker’s own account, Fable 5 is the most powerful model Anthropic has ever released to the public, and it is not close. The company says the system posts state-of-the-art results on nearly every major benchmark used to measure artificial intelligence, with standout performance in software engineering, scientific research, and complex visual and analytical work. The harder and longer the task, Anthropic says, the further ahead Fable 5 pulls from everything that came before it.
To understand why Washington got nervous, you have to understand Fable’s sibling. Fable 5 is the locked-down, public-facing version of a more powerful model called Mythos 5. They are, underneath, the same brain, separated by a layer of safety filters that sit in front of the public version like a bouncer at a door. Anthropic was candid about why it kept the unrestricted version away from the open market. In testing, Mythos reportedly found exploitable flaws in every major operating system and web browser it was pointed at. Rather than hand that to the world, the company stood up a controlled program called Project Glasswing and shared the raw model with roughly fifty vetted organizations, among them Amazon, Apple, Google, Microsoft, and the security firm CrowdStrike, to hunt and patch vulnerabilities before attackers can exploit them.
A Tool That Cuts Both Ways
Here is the uncomfortable part, and there is no polite way around it. A tool powerful enough to find the weaknesses in the world’s software is, by definition, equally useful to the people defending that software and the people attacking it. The capability is neutral. Only the intent is not.
That is the tension the government’s lawyers and the company’s engineers are now fighting over in real time. A model that can audit a codebase and surface every flaw is a gift to a defender racing to patch holes before a breach, and a loaded weapon for someone hunting a way in. The very thing that makes Claude ‘Fable 5’ remarkable is the thing that makes it frightening. On Friday, frightening won.
The Shutdown
The mechanism Washington used is worth pausing on, because it is unusual. The Commerce Department’s Bureau of Industry and Security, in a letter from Commerce Secretary Howard Lutnick to Anthropic chief executive Dario Amodei, did not ban the technology outright. It issued an export-control directive barring access by any foreign national, meaning anyone who is not a U.S. citizen, whether they sit overseas or inside the United States. The order reportedly reaches even Anthropic’s own foreign-born employees. And it appears to go beyond a temporary pause: the directive would require Anthropic to obtain individual government licenses to export, re-export, or even transfer the models inside the country, with potential financial and criminal penalties.
On a shared online service used by people all over the world, there is no clean way to wall off every foreign national on Earth while keeping everyone else online. So Anthropic did the only thing it could to comply. It shut both Fable 5 and Mythos 5 off entirely, for every customer, everywhere. The company’s other Claude models remain fully available and untouched. By several accounts, this is the first time the federal government has reached in and forced a publicly deployed commercial AI model offline. There is no real precedent for what happened Friday night.
Why, and Who Pulled the Trigger
The government’s stated reason is national security. According to Anthropic, the directive did not spell out the specific concern, but the company’s understanding is that officials believe someone discovered a way to “jailbreak” Fable 5, to coax it past its safety filters. Anthropic says it reviewed a demonstration of the technique and found that it surfaced only a handful of already-known, minor software flaws. That is the kind of thing, the company argues, that rival models such as OpenAI’s GPT-5.5 can dig up without any trickery at all, and that ordinary security professionals do every day.
In short, the company says the emergency does not match the response. Anthropic called the order a misunderstanding, said it intends to share more in the coming day, and is working to bring the models back online. Its public objection was blunt: recalling a model used by hundreds of millions of people over a narrow, unproven flaw would, taken to its logical end, freeze every new AI release across the entire industry.
Then there is the question of who set this in motion. An administration official told the news outlet Axios that the government acted after a separate company claimed it could jailbreak Mythos, a claim that rattled officials enough to move. That detail rests on a single, anonymous source and should be read with caution. But if it holds up, it points at something the public record does not yet confirm: that a competitor may have helped knock the most powerful AI on the market offline, three days after it went live.
Have We Hit The Wall on What We’re Allowed to Use?
Strip away the acronyms and the corporate statements, and what is left is a new kind of event. A private company built a tool more capable than anything the public has touched. The public mostly yawned. And the government, treating that same tool as a matter of national security, switched it off overnight using a legal lever almost no one saw coming.
For the millions of people who experience AI as a clever novelty, Friday was a glimpse of how seriously the people in charge are taking it.
The smartest machine ever sold to the public lasted three days before Washington decided the public could not be trusted with it, or that the rest of the world could not. As of this writing, it is still dark, the company is still objecting, and the most important questions are still open. Most that utilize AI, want it back, and promise to be good.
