The US government has lifted the export controls it had imposed on Anthropic's frontier AI models Fable 5 and Mythos 5. The two models had been unavailable worldwide for roughly three weeks starting June 12, but the restrictions were removed as of June 30, and Anthropic says it will begin rolling out Fable 5 globally again from July 1. The episode shows how government involvement in safety has begun to significantly shape how advanced AI is delivered.
Access Resumes After a Three-Week Suspension
Anthropic explained in a blog post that the export controls on Fable 5 and Mythos 5 were lifted as of June 30. Starting July 1, Fable 5 will gradually become available again across the company's own services, including Claude Platform, Claude.ai, Claude Code, and Claude Cowork. In addition, the company plans to restore access through cloud providers such as Amazon Web Services (AWS), Google Cloud, and Microsoft Foundry as quickly as possible.
The controls were put in place on June 12. The measure required Anthropic to restrict access for foreign nationals, but while the order took effect immediately, the company had no reliable way to verify users' nationality in real time. As a result, it was forced to suspend access to both models for all users without being able to narrow down who was affected. As of last week, Mythos 5 had been partially restored for a limited group of organizations within the United States.
US Commerce Secretary Howard Lutnick said the government worked closely with Anthropic in removing the restrictions. In a post on X (formerly Twitter), he said that over the past two weeks or so the government had collaborated with Anthropic to analyze and approve Fable 5, aiming to align positions across the US government and strengthen the country's leadership in AI. The two models had been launched on June 9, meaning the restrictions came just three days later.
Security Concerns Were the Trigger
The controls originally stemmed from a report by Amazon researchers describing a technique that bypassed one of Fable 5's security safeguards. After reviewing the findings together with Amazon and the US government, Anthropic concluded that the technique did not expose any unique Mythos-level cyber capabilities and instead represented a borderline case for Fable 5's safeguards.
Anthropic subsequently retrained the classifier that judges safety, saying the reported technique is now blocked in more than 99 percent of cases. The company also acknowledged that strengthening these protections could increase false positives for some legitimate coding use cases. Researchers at the US Department of Commerce's Center for AI Standards and Innovation (CAISI) also evaluated the updated safeguards.
Frontier AI Has Become "Conditional Infrastructure"
Some experts see the episode as reflecting a structural shift that goes beyond Anthropic alone. Sanchit Vir Gogia, chief analyst at Greyhound Research, said the notable point is not that the restriction was reversed but the framework underlying it. According to him, the US government has applied the long-standing concept of "deemed export" to frontier AI models, signaling a move away from blanket prohibition and toward negotiated, conditional, and monitored oversight.
Gogia described the availability of frontier AI as having become "conditional infrastructure." Noting that a control aimed at foreign nationals ended up shutting off access for everyone because nationality could not be verified in real time, he warned that enterprises cannot assume that spreading a model across multiple clouds insulates them from regulatory actions affecting the model's provider. He suggested that, alongside traditional security, cost, and technical assessments, organizations should also examine risks such as regulatory interruption, cross-border identity restrictions, delays in restoring access, and eligibility as a trusted partner when choosing an AI platform.
The trend of tighter government involvement is not limited to Anthropic. When OpenAI announced a new set of models last week, it said it had shared its rollout plans and the models' capabilities with the US government before launch, and initially limited availability to a small group of trusted partners whose participation had been shared with the government. While the company said it would continue coordinating with the government, it stated that this kind of pre-clearance process should not become the long-term default, reasoning that it keeps advanced AI tools from reaching developers, enterprises, cyber defenders, and partners around the world who need them.
A Call for an Industry-Wide Framework to Assess Jailbreaks
Alongside the announcement, Anthropic called for an industry-wide framework to evaluate AI jailbreaks, techniques that circumvent safeguards. According to the company, there is currently no shared consensus in the industry on how to describe, in objective terms, the severity of a newly discovered technique.
Anthropic said it is working with partners in "Project Glasswing," including Amazon, Microsoft, and Google, to build a framework for assessing jailbreaks. It also plans to expand cooperation with the US government through pre-release testing of future frontier models, information sharing, and joint AI security research. The company argued that if governments are going to be involved in AI releases, there needs to be a durable and transparent process that gives developers and cyber defenders certainty about access to powerful models, and that such rules should be codified in strong regulation and applied equally to all frontier model developers.
Summary
The lifting of export controls on Fable 5 and Mythos 5 shows that the delivery of frontier AI has entered an era shaped not only by technical maturity but also by policy decisions. Restored access does not mean restored certainty for businesses. Going forward, companies will likely need to choose their AI platforms and build their operations on the assumption that regulatory interruptions can happen.
