Anthropic will resume the global rollout of its latest frontier AI model (a cutting-edge, large-scale AI model), Fable 5, in stages beginning July 1. Access had been halted worldwide after the US government imposed export restrictions in June, but those controls were lifted as of June 30. The model will first become available across Claude Platform, claude.ai, Claude Code, and Claude Cowork, with availability on major cloud platforms to be restored afterward.
Fable 5 returns in stages from July 1
Anthropic said in a blog post that the export restrictions on Fable 5 and its sibling model, Mythos 5, were lifted as of June 30. As a result, Fable 5 will roll out worldwide starting July 1, reaching Claude Platform, claude.ai, Claude Code, and Claude Cowork in that order.
For external cloud infrastructure, the company said it will restore availability on Amazon Web Services, Google Cloud, and Microsoft Foundry "as quickly as possible." Its own services come back online first, with access through third-party clouds following behind.
The government was also involved in the reversal. US Commerce Secretary Howard Lutnick said in a post on X that his department had worked closely with Anthropic over the previous two weeks to analyze and approve Fable 5 and to align the effort across the government. Rather than a decision made by the company alone, the resumption came after coordination with authorities.
What happened in June
Fable 5 and Mythos 5 were released on June 9. Just three days later, on June 12, US authorities restricted the export of both models. The stated reason was concern over the models' cybersecurity-related capabilities.
What made the order difficult was that it took effect immediately. The restriction targeted foreign nationals, but Anthropic had no reliable way to verify a user's nationality in real time. As a result, the company had no choice but to suspend access to both models for all users, and a measure aimed at a narrow group effectively became a worldwide shutdown. The outage lasted roughly three weeks, during which only Mythos 5 was partially restored for a limited group of US organizations.
How the restrictions were lifted and safeguards were revised
The episode began with a report from Amazon researchers, who described a technique that could bypass one of Fable 5's cybersecurity safeguards.
Anthropic reviewed the findings together with Amazon and the US government. Their conclusion was that the reported technique "did not expose any unique, Mythos-level cyber capabilities" and represented only a borderline case for Fable 5's safeguards. In other words, it was not judged to be excessively dangerous.
Even so, the company strengthened its protections. It retrained the safety classifier (the mechanism that judges whether an input is dangerous) so that the reported technique is now blocked in more than 99 percent of cases. At the same time, it acknowledged that tightening the safeguards increases false positives for some legitimate coding requests. The updated safeguards were also evaluated by the Center for AI Standards and Innovation, part of the US Department of Commerce.
An era in which availability is shaped by policy
What this episode shows is that whether a frontier model can be used may now be determined not only by its technical maturity but also by policy decisions. For many users, it became an experience of losing access one day and getting it back weeks later.
The same trend extends to other companies. When OpenAI announced GPT-5.6 (Sol, Terra, and Luna) last week, it said it had shared its rollout plans and the model's capabilities with the US government in advance and, at the government's request, initially limited availability to a small group of trusted partners. OpenAI added, however, that it does not believe such government access checks should become a long-term default, expressing concern that powerful tools could be kept from the developers, enterprises, and cyber defenders who need them.
Alongside the resumption, Anthropic also proposed an industry-wide framework for assessing the severity of jailbreaks (techniques that circumvent safety measures). According to the company, the industry currently lacks a common standard for describing the severity of a newly discovered technique in objective terms. To address this, it said it will work with Amazon, Microsoft, Google, and others to build a shared evaluation framework.
Summary
After a worldwide suspension caused by June's export restrictions, Fable 5 will resume availability in stages from July 1. The reversal followed a review that found the reported bypass technique to be limited in scope, along with strengthened safeguards from retraining the classifier. The episode underscores how the availability of frontier AI has become inseparable from policy. The first things to watch are how quickly availability returns across the major cloud platforms, and whether similar temporary suspensions will recur with future model releases.
