Anthropic forced offline by export rules no one can explain

Key Takeaways

• On June 12, 2026, the Trump administration ordered Anthropic to cut access to its newest AI models for all foreign nationals, forcing a global shutdown
• Export control rules have never been applied to AI model access this way, leaving legal experts puzzled about the government's authority
• The incident exposes how unprepared US AI governance is for regulating frontier models without clear, consistent frameworks

Anthropic spent this week scrambling to restore service after the Trump administration ordered the company to block access to its newest AI models, Fable 5 and Mythos 5, for all foreign nationals. The catch: Anthropic could not verify nationality at scale, so it shut down both models for everyone, everywhere. The government has not explained the legal basis for the order.

The directive, dated June 12, 2026, cited "national security authorities" and framed itself as an export control measure. But experts say this application of export rules to AI model access is unprecedented. Export controls traditionally govern physical goods and discrete digital assets, things that cross borders. A chatbot query does not fit that mold.

What are export controls doing in AI chatbots?

Export controls have a long history with weapons, hardware, and sensitive technology. Over time, they expanded to cover software, source code, and technical data. President Biden extended the framework to AI model weights, the core data that allows a model to run independently. The Trump administration dropped that approach. This order goes somewhere else entirely.

When you use Claude, you do not download model weights. You do not receive source code. You send a prompt, Anthropic's servers process it, and you get a response. Nothing tangible crosses a border. The model stays on Anthropic's servers. So what exactly is being "exported"?

Hanna Dohmen, a senior research analyst at Georgetown University's Center for Security and Emerging Technology, told The Verge it remains "an open question" whether the order strains existing rules. "To my knowledge, this is the first time US export controls have been used to control access to an AI model in this way," she said.

Why the government's approach leaves everyone guessing

The administration pointed to a jailbreak potentially used by groups linked to China as partial justification. Anthropic pushed back, saying the jailbreak did not bypass all of its safeguards. But neither party has released the exact language of the directive or clarified which legal authority permits this intervention.

Andrew Reddie, a professor at UC Berkeley's Goldman School of Public Policy, noted that export control and arms regulations give the government "wide latitude" to restrict access. The problem is inconsistency. "The equivocation by successive administrations regarding what the responsibilities of model developers are" has left companies unsure what compliance even looks like, he said.

That uncertainty creates real business risk. If Fable 5 and Mythos 5 triggered this response because of their capabilities, OpenAI, Google, Meta, xAI, and every other frontier lab should expect similar treatment for their next-generation models. If the order targeted specific safeguard gaps, the government has not explained what protection would be sufficient. And if Anthropic was singled out because of its testy relationship with the administration, the precedent becomes even murkier.

What does this mean for other AI companies?

The short answer: nobody knows. Congress is already trying to close the loophole around remote access to cloud services. Legislation is moving through the Senate. But until that passes, the rules are murky, and the executive branch appears willing to act unilaterally.

Discussion on Hacker News and similar platforms has centered on what some call "government kill-switches" in commercial software. If Washington can order an AI company to shut down global access overnight, without public legal justification, the precedent extends far beyond Anthropic. Any company building frontier models now operates in a regulatory environment where the rules can change instantly and opaquely.

International customers are taking note. The incident adds fuel to arguments that governments and companies outside the US should be wary of depending on American AI providers. If access can be revoked without warning, building critical infrastructure on top of these models becomes a strategic liability.

Is this sustainable AI governance?

Experts say no. "This episode makes clear the unsustainability of the existing governance regime," one researcher told The Verge. Governing AI through ad hoc interventions, with opaque legal justifications, does not scale. It creates uncertainty for companies, erodes trust with international partners, and fails to establish the kind of clear framework that would let the US maintain its lead in AI development.

The Biden administration tried to create rules around model weights. The Trump administration discarded them and reached for a different tool. Neither approach has produced a coherent policy. Frontier AI is moving faster than the regulatory apparatus, and the gap is widening.

“To say that this is an unsettled area of export control rule-making would be an understatement.”

— Andrew Reddie, UC Berkeley Goldman School of Public Policy

Frequently Asked Questions

Why did Anthropic shut down Fable 5 and Mythos 5 globally?

The Trump administration ordered Anthropic to block access for all foreign nationals. Because Anthropic could not verify user nationality at scale, it disabled the models for all users worldwide to ensure compliance.

What legal authority did the government use?

The administration cited "national security authorities" and described the order as an export control directive. It has not publicly released the precise legal language or explained how export rules apply to remote AI access.

Does this affect other AI companies like OpenAI or Google?

Potentially. If the order targeted Anthropic's model capabilities, similar restrictions could apply to next-generation models from any frontier lab. The government has not clarified its criteria.

Are there any proposed laws to address this gap?

Yes. Congress is working on legislation to bring remote cloud service access under export control rules. The bill is currently moving through the Senate.

What was the government's stated concern?

The administration referenced a jailbreak potentially used by groups linked to China. Anthropic disputed the severity, saying the jailbreak did not bypass all safeguards.