Fable 5: The AI So Powerful the US Government Shut It Down in 90 Minutes

Image Credit: Nicolas Tucat/Agence France-Presse — Getty Images

Two days ago, I had a discussion with my friend Amara who told me something that I found really interesting.

It was the evening of June 12, 2026, and the remote-worker software developer had spent the better part of three days working alongside Claude Fable 5 on her company’s most ambitious project yet: a full migration of a complex financial application from one programming language to another. The kind of task that would normally take a team of experienced engineers several weeks. Fable 5 was doing it in hours. Writing its own tests. Catching its own errors. Moving autonomously through thousands of lines of code with a precision that felt, to Amara, less like using a tool and more like working alongside a gifted colleague who never tired and never lost focus.

Then, at 5:21pm Eastern Time in the United States, everything stopped.

No warning. Just a message: access to Claude Fable 5 had been suspended.

Amara was not alone. Across the world, developers, researchers, and analysts who had spent the previous four days discovering what this new model could do all received the same abrupt message. Fable 5, the most powerful AI model ever made available to the general public, had been switched off. Not because of a technical failure. But because the United States government had ordered it gone.

A New Kind of Intelligence

To understand why Washington acted so dramatically, you first need to understand what Fable 5 actually is and why its release on June 9, 2026 felt like a genuine inflection point.

Anthropic, the San Francisco-based AI safety company behind the Claude family of models, had spent months carefully preparing the world for what it called its “Mythos-class” tier of intelligence. The name itself, Mythos, from the Greek for story or legend, was a signal that something categorically different was coming.

Mythos-class models represent a tier of Claude models that sit above the previous Opus class in capability. They are not simply faster or more knowledgeable than what came before. They represent a different order of capability entirely.

Claude Fable 5 is Anthropic’s most capable model for ambitious coding projects, including multi-day autonomous sessions. It can write its own tests to check its work, implement designs with high fidelity, and use vision to check outputs against goals.

For non-technical readers, consider what that means in practice. Previous AI models required constant human supervision. You would give them a task, review what they produced, correct mistakes, give the next instruction, and repeat. Fable 5 changed that relationship fundamentally. Both Fable 5 and Mythos 5 can run “unattended” for longer stretches than previous Claude models, carrying out sequences of instructions without constant supervision.

Real-world users confirmed this almost immediately. One developer tested Fable 5 against a benchmark prompt describing a financial planning application with twenty forms and complicated financial simulation mathematics. Every model previously tested had failed the task. Fable 5 completed it in thirty minutes.

The model is state-of-the-art on nearly all tested benchmarks and delivers exceptional performance in software engineering, knowledge work tasks, and vision, built for ambitious, long-running work. In short, this was not an upgrade.

The Capability That Frightened Washington

But there was one capability in Fable 5 that made governments deeply uncomfortable.

Fable 5 was extraordinarily good at finding security vulnerabilities in software.

This might sound like a narrow technical concern, but its implications are enormous. Almost everything that runs modern civilisation, from banks to hospitals to power grids, runs on software that still contains hidden flaws. If discovered, those flaws can be exploited to cause catastrophic damage.

For decades, finding these flaws required teams of highly skilled human security researchers working painstakingly through code. Fable 5 could do it faster and more thoroughly than any human team.

Since April, when an earlier Mythos preview was sent to about 150 organisations under the banner of Project Glasswing, users had already reported more than 10,000 critical security flaws in their own systems. Used defensively, to find and fix vulnerabilities before malicious actors could exploit them, this capability was invaluable. But the same capability in hostile hands could be turned into an offensive weapon of unprecedented power.

Anthropic understood this tension and built guardrails accordingly. Fable 5 runs on the same underlying model as Mythos 5, but with hard limits built in. The system is designed to refuse or redirect requests related to cybersecurity, biology, and chemistry. When those guardrails trigger, the query is silently routed to an older model, Claude Opus 4.8, instead.

The company had also stress-tested those guardrails before launch. Anthropic ran an external bug bounty that produced no universal jailbreaks in over 1,000 hours of testing, and then worked with external red-teaming organisations which also failed to find universal jailbreaks.

They believed the model was safe enough to release. Washington was about to disagree.

90 Minutes to Comply

On Friday, June 13, 2026, the US Commerce Department issued an emergency export control directive banning Claude Fable 5 and its underlying model Mythos 5 from use by all foreign nationals, whether inside or outside the United States.

Anthropic received the directive at 5:21pm Eastern Time. The letter did not provide specific details of its national security concern. The company was given just 90 minutes to comply.

The government’s concern centred on a reported jailbreak, a method of bypassing Fable 5’s safety guardrails to extract information about software vulnerabilities. The scope of the directive was sweeping. It applied not just to people outside the United States, but to any foreign national anywhere in the world, including Anthropic’s own non-citizen employees. The net effect was that Anthropic had to abruptly disable Fable 5 and Mythos 5 for all customers to ensure compliance.

The reaction was immediate and divided. Pentagon officials framed the action as a straightforward national security measure. But cybersecurity experts pushed back hard. Anthropic disputed the action, arguing the jailbreak capability identified was not unique to its model. Katie Moussouris, a respected cybersecurity researcher who reviewed the jailbreak report, was blunt: the capabilities Amazon’s researchers had identified were exactly what security defenders need. “The administration got it wrong,” she said.

Critics also noted a glaring inconsistency. The government freeze did not apply to leading models from Anthropic’s competitor OpenAI, despite those having demonstrated comparable capabilities.

A Feud Hiding in Plain Sight

To those watching closely, the Fable 5 shutdown was not simply about a jailbreak. It was the most visible eruption of a conflict that had been building for months.

Anthropic is suing the Trump administration after being put on a supply chain blacklist for its refusal to allow the US military to use its AI models for domestic surveillance and fully autonomous weapons systems. The company had consistently positioned itself as the AI safety conscience of Silicon Valley, the firm most willing to say publicly that these systems posed real risks and needed real guardrails. That stance had earned it fierce antagonism in parts of Washington.

Administration officials and advisers had accused the company of doomerism, warning of AI risks and stoking fear, then continuing to release more powerful models anyway.

Now, just as Anthropic was preparing to file for a public stock market listing, the administration had found its most powerful lever yet.

What Comes After Fable

Here is what history teaches us about this moment.

When OpenAI released the first version of ChatGPT in late 2022, the world was astonished. Millions of people who had never thought much about artificial intelligence were suddenly having conversations with a machine that felt genuinely intelligent. It felt like the future arriving ahead of schedule.

But today, looking back at those early ChatGPT conversations, the limitations are obvious. The model was prone to confident errors, struggled with complex reasoning, and had almost no ability to take autonomous action. What felt revolutionary in 2022 looks, from the vantage point of 2026, like a rough early sketch.

Fable 5 will look the same way one day. The models that follow it will be more capable and more deeply woven into the infrastructure of daily life. The questions being fought over today, who gets to use them, what safeguards are necessary, which governments have the authority to switch them off, will only grow more urgent.

Amara eventually found a workaround and finished her project using older tools, taking three times as long. She was frustrated but philosophical. “The model was real,” she said. “What it could do was real. They can suspend access. They cannot unsee what we all saw.”

She is right. The genie does not go back into the bottle.

The Fable 5 shutdown is not the end of a story. It is the opening paragraph of a much longer, far more consequential one — the story of how humanity learns to live with intelligence that increasingly exceeds its own.