Tragic tale of Claude Fable, the most powerful AI ever created

For three days in June, the most powerful artificial intelligence ever offered to the public belonged to everyone who had a Claude subscription.

Then it was gone.

Not crashed. Not recalled for a bug. Switched off, on the written order of the United States government.

On June 9, Anthropic, the San Francisco lab behind the Claude chatbot, released Claude Fable 5 and called it the most powerful model it had ever made available to the general public. On 12 June, just three days later, a letter from the US Commerce Department arrived. Reportedly, the company was given 90 minutes to comply.

The order, citing national security, barred Fable, and the already-restricted Mythos 5, from any foreign national, whether inside or outside the United States, down to Anthropic’s own foreign-born staff.

The company could not cleanly wall off Americans from everyone else. So, it did the only thing it could, and pulled Fable from every customer on earth.

Read that again. A product used by millions went dark worldwide because a government decided who was not allowed to touch it.

To feel the shock, you have to understand what Fable 5 actually is. It was state of the art on almost every measure that matters, from writing software to scientific research, and it could grind through long, multi-step problems that defeated every model before it.

Strip away the branding and Fable is Anthropic’s most powerful system, Mythos, wrapped in safety rails and handed to the public for the first time. Mythos itself has never been on open sale. It goes only to a handful of approved cyber defenders, critical infrastructure operators and governments.

In other words, the public never had the raw article. Fable was as close as the rest of us were ever going to get. And now even that is gone.

This is the part that should unsettle you. It appears to be the first time a government has reached into a commercial AI lab and forced a live, public model offline. Frontier AI has just been treated like a fighter jet or a microchip, a strategic export to be switched on and off at a government’s pleasure.

The stated trigger was almost comically small. Anthropic says officials pointed to a narrow “jailbreak”, reportedly flagged by Amazon, that nudged the model into finding software flaws.

The company insists that rival models do the same thing, that OpenAI’s GPT-5.5 is no different, and that network defenders use exactly this capability every day to keep systems safe.

Anthropic is upset, calling the decision a “misunderstanding” and stating it could not see why “a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people”.

Even Washington’s own people are baffled. Dean Ball, who helped write the Trump administration’s AI Action Plan, said he could not tell “if this is lawfare against Anthropic in particular or extreme national-security hawkery”, and called the order cartoonish. His sharpest jab landed on the contradiction: the same government has been loosening sales of advanced AI chips to China, even as it bars citizens of close allies from opening an American chatbot.

For everyone outside America, the lesson is cold and simple. The most capable intelligence ever built can be granted or revoked at the stroke of one capital’s pen.

Europe felt it at once. Politicians called the blackout a “wake-up call”, warning that the continent “cannot settle for being an open market dependent on technologies designed, funded and controlled elsewhere”. If Berlin and Paris feel exposed, imagine the view from Kuala Lumpur.

This is how a two-tier world is built. A few states and companies own the frontier and the chips that train it. Everyone else rents, on terms that can vanish in 90 minutes. Capability, not just wealth, becomes the line between the powerful and the rest.

So where does that leave Malaysia? Let us be honest, more honest than the usual talk of “sovereign AI” allows.

We are not going to build a model like this. Not this year, not this decade. We do not have the capital, we do not have the concentration of talent, and, if we are honest, we do not have the national appetite for it.

These models are built by companies that employ the leading minds in the world and that raise tens of billions of US dollars, hundreds of billions of ringgit, in a single funding round. That is not defeatism. It is arithmetic.

So the task is not to pretend we will one day out-build Anthropic. It is to stop behaving as if these tools will always be there. We should spread our bets across American, Chinese and open-source models, and keep anything that truly matters—in government, in banking, in security—running on systems no foreign official can switch off from the other side of the planet.

Fable 5 may be back within days, and Anthropic says it is trying. But the lesson will outlast the outage.

The most important tools of this century now sit behind a switch. The switch works. And our hand is nowhere near it.

The writer can be contacted at kathirgugan@protonmail.com.

The views expressed are those of the writer and do not necessarily reflect those of FMT.