If you opened Claude this week, you'll have seen the popup: Fable 5 is back.
It's a strange little banner to sit at the centre of one of the more dramatic stories in AI this year, because the model it's welcoming back was pulled offline by the US government less than a week after launch.
A quick recap for anyone who missed it.
Anthropic launched Claude Fable 5 on June 9 alongside Mythos 5, a less restricted sibling reserved for vetted cybersecurity partners. Fable was the public-facing version, and early users were impressed enough that plenty of people signed up for Claude subscriptions purely to try it.
Then, on June 12, the US Commerce Department issued an export control directive after Amazon researchers reported a jailbreak that got Fable to identify software vulnerabilities and, in one case, produce exploit code. The order banned access for any foreign national, and since Anthropic had no way to verify nationality in real time, it simply switched the model off for everyone.
Three days of availability, then nothing.
Anthropic's position throughout was that the fuss was overblown, arguing that weaker models, including its own Opus 4.8 and OpenAI's GPT-5.5, could find the same vulnerabilities and produce the same exploit demonstration. The government disagreed strongly enough to keep the model dark for nearly three weeks while the two sides negotiated.
The fix that broke the deadlock was a new safety classifier trained to catch the reported technique, which Anthropic says now blocks it in over 99% of attempts. Flagged requests get rerouted to Opus 4.8, and Anthropic has warned that some routine coding and debugging work may trip the filter too.
So Fable is back as of July 1, across Claude.ai, the API, Claude Code and Cowork, with cloud availability on AWS, Google Cloud and Microsoft to follow. The catch is in the small print. Pro, Max, Team and select Enterprise users get Fable included for up to 50% of their weekly usage limits until July 7, after which it moves to usage credits.
That's a six-day window at half capacity, replacing what was originally full inclusion. It also burns through usage faster than Opus, so the ceiling arrives sooner than you'd think.
For founders, the more interesting story is what the blackout revealed. Companies that had built workflows around Fable lost access overnight, with no appeal and no timeline. Rivals used the gap, with OpenAI's GPT-5.5-Cyber topping a Berkeley security benchmark while Anthropic's models sat on the bench.
If your product or your operations lean on a single frontier model, this was a live demonstration that a government directive can remove it on a Friday afternoon. The model is back. The precedent isn't going anywhere.
