Three days of Fable: the rise and sudden halt of Anthropic's most powerful model

Some models live for years. Fable 5 lived for three days. On June 9, 2026 Anthropic positioned it as the strongest Claude model ever made generally available; by June 12 at 5:21 PM ET it was switched off again. Not because Anthropic pulled it, and not because the model misbehaved, but because the US government issued an export control directive. This is the story of those three days, and what it means for everyone who works with Claude.

Day one: a new class above Opus

Fable 5 was not an ordinary successor. It belonged to the so-called Mythos-class, a model tier that Anthropic says sits above the Opus class in capability. Two models on the same underlying base appeared on the same day: Claude Fable 5, the public version with full safety measures, and Claude Mythos 5, the exact same model with the cyber safety classifiers removed, available only to approved cyber defense partners through Project Glasswing.

The name was chosen carefully. "Fable" comes from the Latin fabula, "that which is told", related to the Greek mythos. The only real difference between Fable and Mythos was the safety layer, not the brain underneath.

The price underlined the ambition: 10 dollars per million input tokens and 50 dollars per million output tokens. That sounds steep next to Opus 4.8 ($5/$25), but Anthropic described it as "less than half the price" of Mythos Preview, the very first Mythos-class model from April, which cost $25/$125.

What Fable 5 could do

The benchmarks were the kind of numbers you build a launch event around. On SWE-bench Pro, Fable 5 scored 80.3 percent, against 69.2 percent for Opus 4.8, a lead of more than eleven points on exactly the terrain, agentic coding, where the difference matters in practice. On SWE-bench Verified it reached 95.0 percent. On GDPval-AA and on Cognition's FrontierCode it ranked first.

The most striking example came from Stripe. The company used Fable 5 to migrate a 50-million-line codebase in a single day, work that a team of engineers was estimated to need two months for. That is exactly the kind of long, complex, multi-day task a Mythos-class model was built for.

Day three: the directive at 5:21

On Friday June 12, 2026 at 5:21 PM ET, Anthropic received a directive from the US Commerce Department, signed by Secretary Howard Lutnick and co-drafted by the Bureau of Industry and Security. The order prohibited access to both Fable 5 and Mythos 5 for, in Anthropic's own wording, "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees".

Anthropic cannot filter on nationality in real time. The result: to comply with the order, the company had to disable both models worldwide for all users. According to NBC News and Bloomberg, this is likely the first time a leading AI company took a publicly deployed model offline due to direct federal intervention.

The disputed reason

The letter gave no concrete details about the national security concern. Anthropic's understanding is that the government believes a method exists to "jailbreak" Fable 5: a technique where you ask the model to read a specific codebase and fix software flaws.

Anthropic strongly disputes its significance and calls it a misunderstanding. The demonstrated technique, according to the company, yielded only a small number of previously known, minor vulnerabilities, exactly the kind defenders search for daily, and which other public models such as OpenAI's GPT-5.5 find just as well without any bypass. Fable 5 was also designed with a "defense in depth" strategy: making jailbreaks narrow or expensive, combined with thorough monitoring and a mandatory 30-day data retention policy for rapid detection.

Independent expert Katie Moussouris, CEO of Luta Security, was shown the report and called the government's response, to the Wall Street Journal, "a complete overreaction because this is exactly the kind of prompting that defenders would do". According to WSJ reporting, the trigger came from Amazon, where researchers reportedly jailbroke Fable 5 and escalated their findings to the government. Worth noting: that origin story rests largely on anonymous sources and has not been officially confirmed by Amazon.

The bigger picture

The suspension did not come out of nowhere. The relationship between Anthropic and the Trump administration was already tense: earlier in 2026 the Pentagon labeled Anthropic a "supply chain risk", after which Anthropic filed a lawsuit that is still pending. At the same time the company was in an upswing, a Series H of 65 billion dollars at a 965 billion post-money valuation, run-rate revenue that crossed 47 billion earlier in the month, and a confidentially filed IPO prospectus.

So this is not a permanent "ban" and certainly not a punishment for a malicious model. It is a disputed suspension on national security grounds, which Anthropic wants reversed. An administration official suggested possible restoration "in the next few weeks", but there is no fixed date.

What this means for your work

The good news first: all other Claude models remain available. Only Fable 5 and Mythos 5 are suspended.

Use Claude Opus 4.8 now. It is the most powerful model currently available, and exactly the fallback Fable 5 routed to. For a balance of speed and cost, pick Sonnet 4.6; for volume and cheap processing, Haiku 4.5.

Don't count on Fable 5 returning soon. Treat it as unavailable until Anthropic or the government announces restoration. Also watch the subscription transition that was scheduled for June 23, Fable 5 was set to move off Pro, Max, and Team plans and shift to usage credits, that plan has been thrown into doubt by the suspension.

Follow the official channels. The situation moved day to day around June 12 to 14. For the current status, anthropic.com/news and the model list at platform.claude.com are authoritative.

What three days of Fable teach us

Fable 5 briefly showed what a Mythos-class model is capable of: a 50-million-line codebase in a day, an eleven-point lead over the strongest public model before it. But it also showed something else, that the stakes of frontier AI are now high enough for a government to step in within three days, even when that intervention is technically contestable.

For anyone working with Claude every day, little changes in practice: Opus 4.8 is ready and does the job. But the short life of Fable 5 is a useful reminder that the most powerful models are now shaped not only by their makers, but by the world around them.