Claude Fable 5 was made available globally again on July 1. Almost three weeks after the U.S. government ordered Anthropic to disconnect the model for users outside the country, the Commerce Department lifted the directive and the company reactivated Fable 5 with a new safety classifier and transparency mechanisms.
The return is not a simple restoration of service. It is a public negotiation between safety, government control, and commercial access.
What changed
According to Anthropic’s announcement, Fable 5 returns with three main modifications:
- An additional safety classifier, trained specifically to detect the behavior described in the Amazon report that motivated the Commerce Department directive. Anthropic claims the classifier has a block rate above 99% on that target behavior.
- Greater transparency on blocked interactions. When a request triggers the safeguards, the user receives an explicit notification and is temporarily reverted to Opus 4.8 for that query.
- Subsidized access for one week, until July 7, as a concession to users affected by the shutdown. After that, normal usage credits apply.
Mythos 5, on the other hand, remains limited to Glasswing partners and was not reactivated alongside Fable 5.
The precedent that does not disappear
On June 12, the U.S. government ordered Anthropic to suspend global access to Fable 5 and Mythos 5. The official reason: a jailbreak method that, according to Washington, could compromise national security. Anthropic complied immediately but publicly argued that the jailbreak was “narrow and not universal” and that similar capabilities existed in other market models.
The controversy was not resolved legally. What happened was a technical adaptation: Anthropic built the safeguard the government demanded and demonstrated that it could be implemented without completely disconnecting the model.
This establishes a pattern. If a government can demand the disconnection of a model for a specific vulnerability, and the provider can mitigate that vulnerability with a classifier, the future cycle becomes: discovery → sanction → patch → reactivation.
Sonnet 5, in parallel
On the same day Fable 5 returned, Anthropic introduced Claude Sonnet 5 as its new mid-range offering. Through August 31, its promotional price is $2 per million input tokens and $10 per million output tokens. After that, it moves to $3 and $15.
The move matters: with Fable 5 reserved for frontier tasks and Opus 4.8 as fallback, Sonnet 5 completes the lineup. For teams that need to distribute load between depth and cost, there is no longer an excuse not to diversify within Anthropic’s catalog.
What it means for users
For developers and companies that had integrated Fable 5, the return is an immediate operational improvement. But it is also a warning. Dependence on a model that can be disconnected by regulatory decision does not disappear because the model has returned. What changes is that there is now a documented procedure for its suspension and reactivation.
The Fable 5 case remains the clearest example of how AI geopolitics no longer applies only to chips and data: it now also applies to active models, connected users, and decisions made in real time.
Related reading
- Anthropic — Redeploying Fable 5
- Anthropic — Introducing Claude Sonnet 5
- The Information — report on OpenAI’s inference cost optimization
- IA al Día — The US Government Ordered Claude Fable 5 Shut Down