Anthropic launched its most ambitious AI models on a Monday. By Thursday, the US government had pulled the plug on global access.

The Commerce Department issued an export control directive on June 12, 2026, ordering Anthropic to immediately suspend access to Claude Fable 5 and Mythos 5 for all foreign nationals. That includes Anthropic’s own employees who aren’t US citizens. The directive, signed by Commerce Secretary Howard Lutnick at 5:21 p.m. ET, came just three days after both models went live on June 9.

Anthropic complied that same evening, disabling both models for all users globally. The company called the whole situation a “misunderstanding.”

What triggered the shutdown

The core dispute centers on a jailbreak. Someone found a way to bypass Fable 5’s safety guardrails, and what that unlocked apparently spooked the government.

Anthropic described the vulnerability as a narrow issue. US officials saw it differently, arguing the jailbreak could grant access to advanced cybersecurity capabilities tied to the Mythos model. The government believed someone could exploit the safety hole to weaponize features that were supposed to stay locked down.

Amazon reportedly identified the jailbreak concerns first and flagged them to the government. That detail matters because Amazon is Anthropic’s largest investor and cloud infrastructure partner.

The Trump administration claimed it had previously warned Anthropic CEO Dario Amodei about the risks. Anthropic has pushed back on the severity of those earlier warnings.

The scope and the stakes

The export control directive is narrow in one important sense. Only Fable 5 and Mythos 5 are affected. Every other Anthropic model remains accessible to users worldwide. The company has indicated it intends to restore access to both models once it adequately addresses the government’s concerns.

The directive’s language bars access to “any foreign national, including Anthropic employees,” meaning the company can’t even let its own international engineers work on the affected models.

Fable 5 and Mythos 5 were designed as safeguarded versions of Anthropic’s frontier capabilities, built specifically for broad availability. Anthropic has positioned Fable 5 as its most sophisticated AI model available to the public, leveraging the same foundational technology as the restricted Mythos 5, and equipped with additional safety classifiers intended to mitigate risks in sensitive applications such as cybersecurity.

Broader context for the AI industry

Anthropic has spent years positioning itself as the safety-first AI company. Dario Amodei and his team left OpenAI specifically because they wanted to build AI with stronger guardrails.

Export controls have traditionally targeted hardware, like the chip restrictions aimed at China. Applying them to a software model, and doing so within 72 hours of launch, represents a meaningful expansion of the government’s toolkit.

What this means for investors

Anthropic is privately held, so there’s no public stock to tank. But the ripple effects touch publicly traded companies across the sector.

Amazon, as Anthropic’s primary cloud partner and major investor, faces indirect exposure. If Anthropic’s ability to deploy frontier models gets entangled in export control disputes, that affects the value proposition of the partnership.

The fact that Amazon was reportedly the entity that flagged the vulnerability to regulators suggests that even within AI companies’ own ecosystems, there are multiple stakeholders with different risk tolerances and different incentives around disclosure.