Temporary Unavailable

And probably back soon.

Anthropic has spent the past few months telling anyone who would listen that Claude Mythos is extremely powerful and potentially dangerous. It launched Mythos only to a small circle of trusted organizations through Project Glasswing, then built an elaborate set of safety classifiers (detailed in the announcement) into Fable 5, the public version, to silently reroute flagged queries to an older model before anyone could do anything interesting with them. The categories that triggered the rerouting were offensive cybersecurity, dual-use biology and chemistry, and attempts to extract Fable 5 capabilities to train competing models. Anthropic put a lot of effort into publicly communicating that the underlying model was, in their estimation, quite scary.

The Trump administration believed them! Or at least, someone did. On June 12th, three days after the launch, Commerce Secretary Howard Lutnick sent a letter to Dario Amodei directing Anthropic to suspend access to both models for all foreign nationals, citing national security authorities. The stated reason was a jailbreak that could unlock the cybersecurity capabilities Anthropic had been carefully routing around.

The tip came from an unnamed competitor. The notice was verbal only. No written technical evidence. Anthropic has no nationality verification system, so an order targeting foreign nationals produced, in practice, a global shutdown for everyone.

By that evening, Claude’s landing page read: Fable 5 is temporarily unavailable. This appears to be the first time a government has forced a publicly deployed frontier AI model offline.

The consistency problem is the part I keep coming back to. GPT-5.5, which Anthropic says has the same cybersecurity capabilities, was not touched. The DoD had already classified Anthropic as a supply chain risk to national security in February (a label usually reserved for Huawei), after Anthropic refused to let Claude be used for mass domestic surveillance and fully autonomous weapons. Anthropic filed suit.

Trump later wrote on Truth Social:

Anthropic said no amount of intimidation would change its position. The Fable 5 shutdown arrived five months into that fight, triggered by a tip from an unnamed competitor whose identity nobody has published. Someone should probably ask who.

Anthropic said it plainly:

We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

I think this is right. Every frontier model can be jailbroken given sufficient effort. Any competitor can report a jailbreak to a Commerce official. If claimed jailbreaks are sufficient grounds for recall, you have handed a veto over AI development to anyone willing to run a technique and make a phone call. Whether the underlying capability concern is legitimate (to be fair, Fable did seem kinda scary with the little time I had with it) does not change much about how the mechanism here looks.

There is also quite a bit of irony to the classifier situation. Fable 5 launched with classifiers that silently rerouted suspicious queries to Opus 4.8, the older model that remains available. Users almost immediately noticed the classifier triggering on routine greetings. Anthropic apologized and adjusted it on June 11th. The government issued the shutdown order the following evening. Anthropic built safety infrastructure into Fable 5 specifically because Mythos is dangerous enough to require it, and that safety infrastructure publicly quantified just how dangerous Mythos was.

---

Onto the markets! There are a lot of them.

The most liquid is “Will Fable be reenabled for Americans before June 20?” at 17%

“Will Fable be reenabled for Europeans before July 1?” has moved dramatically since Anthropic’s Managing Director of International Chris Ciauri told a Seoul press conference on Wednesday that they were “very confident that in the coming days, the models will become available again.” That market has gone from 22% to 57%, up 35 points.

The prop bets market on how restoration looks has also shifted. “Fable available with safeguards more stringent than previously, even for Americans” leads at 50%, down from 57%. The two-tier scenario is at 17%.

Separately, someone has created a market asking whether export controls on Fable 5 will be rescinded within 72 hours of GPT 5.6’s release, which sits at 26%. The competitive pressure theory, in other words, has its own market now.

---

I should mention that Anthropic had just confidentially filed for an IPO when all of this happened, because there is a theory circulating that deserves at least acknowledging. The narrative the shutdown generates, a model so powerful the US government shut it down for national security, is extraordinarily good IPO marketing.

You cannot buy that kind of capability signal. But IPO pricing is about long-term narrative, not a week of token movement, and the administration is so afraid of Claude that they forced it offline is a better founding story than almost anything Anthropic could have written themselves. Whether the shutdown was anticipated, welcomed, or genuinely unwelcome is something only Anthropic knows, but the incentive structure is at least worth being aware of.

What comes after Mythos in the model naming sequence? Haiku, Sonnet, Opus, Fable, Mythos.

There is a market on this, appropriately. Nobody currently knows, and at the moment nobody can access Mythos to ask it.

Happy Forecasting!

- Above the Fold