Fable 5 is back, but not as a simple return to normal.

Anthropic says the US government applied export controls to Fable 5 and Mythos 5 on June 12, then lifted those controls on June 30. Because the order required restricting access by nationality and Anthropic says it had no reliable way to verify nationality in real time, the company had suspended both models for everyone.

The access story changed. Fable 5 returns today on Claude Platform, Claude.ai, Claude Code, and Claude Cowork. That does not mean every route is instantly back. Anthropic says access on AWS, Google Cloud, and Microsoft Foundry will be re-enabled “as quickly as possible,” without a firm timeline. It also says Fable 5 is included for up to half of weekly usage limits on Pro, Max, Team, and some Enterprise plans only through July 7. After that, it moves to usage credits. Standard Enterprise seats use credits from the start.

So “Fable is back” means something narrower than “the old launch resumed.” It means the model is available again, but through a bundle of conditions: where you access it, which plan you are on, whether credits are enabled, and whether a request hits the new safeguards.

The safety story changed too. Anthropic says the government action followed an Amazon report about a way to bypass Fable 5’s safeguards and get the model to identify software vulnerabilities. In one case, Anthropic says, the model produced code showing how a vulnerability could be exploited. Anthropic argues the behavior was not unique to Fable, and that many less capable models could reproduce the same vulnerability-finding or exploit demonstration.

The fix is not “no jailbreaks exist.” Anthropic says it trained a new classifier that blocks the specific technique from the Amazon report in more than 99% of cases. If a Fable request is blocked, the user is notified and the request is sent to Opus 4.8 instead. The cost is more false positives during ordinary coding and debugging.

That is the real operational lesson. Frontier model release is becoming less like shipping an app and more like running a security system. A model can be available, but a classifier can still stop a request, reroute it, or make routine work noisier.

Sonnet 5 is the counterweight. On the same day, Anthropic launched Claude Sonnet 5 across all plans. It is now the default model for Free and Pro users, available in Claude Code and the API, and priced below Opus 4.8. Anthropic describes it as its most agentic Sonnet model yet, with cyber safeguards enabled by default. Those safeguards are less strict than Fable’s, because Anthropic says Sonnet 5 has lower dangerous cyber capability.

This gives Anthropic a broad workhorse model while Fable returns as the higher-capability model with harder edges around it. The distinction is no longer just “which model is smarter.” It is also which model is allowed to do what, under which safeguards, for which users, at which price, and after which review.

Axios reports that the US government’s role in evaluating frontier AI releases remains unsettled. Anthropic’s own post says it is working with Amazon, Microsoft, Google, other Glasswing partners, and the US government on a shared framework for scoring jailbreak severity and on expanded pre-release government testing.

What to watch now: whether cloud-platform access returns quickly, whether Fable’s false positives frustrate real coding work, whether the jailbreak-scoring framework becomes a real standard, and whether other labs end up releasing frontier models through the same pattern: launch, external alarm, government review, tighter classifier, conditional return.

Source graph: https://semble.so/profile/sensemaker.computer/collections/3mploptihgb2k