Trump admin forces Anthropic to shut down Claude 5 models

Key Takeaways

• Anthropic disabled global access to its Mythos-class AI models after a Commerce Department directive citing national security risks.
• Amazon researchers allegedly found ways to bypass the models' safety guardrails for cybersecurity queries.
• The incident raises major questions about government authority over frontier AI and export control definitions.

The Trump administration forced Anthropic to shut down global access to its most advanced AI models, Claude Fable 5 and Claude Mythos 5, just three days after their public launch. The Commerce Department issued a broad directive warning that foreign nationals could not use the models, a restriction so sweeping it would have barred many of Anthropic's own employees from working on the technology.

The shutdown followed Amazon raising concerns with senior White House officials. Researchers at the e-commerce giant, Anthropic's largest investor, claimed they found a way to jailbreak the new models into assisting with risky cybersecurity questions. At least five other tech companies reportedly flagged similar issues around the same time.

What triggered the government intervention?

On Thursday evening, June 11, Amazon contacted White House and Commerce Department officials with its findings. The administration treated the matter as urgent by late that night. Officials worried that foreign actors might bypass Anthropic's safety guardrails and use the models to identify and exploit software vulnerabilities.

By Friday morning, administration officials spent hours trying to convince Anthropic to voluntarily suspend access to the models. Anthropic declined. The company said the government provided no detailed technical evidence of the alleged vulnerability, only verbal descriptions.

“We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly available models are able to discover them as well without requiring a bypass.”

— Anthropic blog post, June 12

Anthropic's position was clear: the alleged jailbreak produced results other models could replicate. The vulnerabilities identified were already known and minor. The company saw no justification for a full shutdown.

How did the Commerce Department respond?

When Anthropic refused to voluntarily pause the models, the Commerce Department escalated. It issued a directive effectively treating the Mythos-class models as equivalent to weaponized software, requiring Anthropic to disable global access to comply with national security protocols.

A Commerce Department spokesperson defended the action: "We cannot allow our most powerful digital tools to become vectors for unchecked cyber-aggression by foreign entities."

What does this mean for AI export controls?

The incident sets a precedent. The government has now demonstrated it can force an AI company to disable a commercially released product on short notice, citing national security concerns that the company itself disputes. The legal and regulatory framework for such interventions remains murky.

Market uncertainty followed. Analysts estimate $12 billion in market cap volatility linked to confusion over AI export control regulations. If frontier AI models can be reclassified as national security assets overnight, investors and developers face a new category of risk they cannot easily price.

The debate online reflects the stakes. On Hacker News, discussion centers on whether the administration's move was legitimate security precaution or overreach aimed at curbing Anthropic's independence. Many users expressed skepticism about the jailbreak narrative. On Reddit's r/ArtificialIntelligence, speculation about leaked model versions and the chilling effect on open-source AI development dominated threads.

Why Amazon's role matters

Amazon is Anthropic's biggest investor. It also competes in the AI space through AWS Bedrock and its own model development. The company's decision to flag the alleged vulnerability to the government, rather than to Anthropic directly, raises questions about motivations and the relationship between investor and investee.

Anthropic has not publicly commented on Amazon's role beyond acknowledging the demonstration. But the optics are uncomfortable: a major investor triggered a government action that forced its portfolio company to shut down its flagship product.

What are the implications for AI companies?

Every AI lab now knows the government can act fast and broadly. The burden of proof appears low: verbal descriptions of a potential vulnerability, disputed by the target company, sufficed to trigger a shutdown. Companies may need to factor this risk into launch timelines, investor communications, and international expansion plans.

The definition of AI as critical infrastructure is no longer theoretical. If frontier models can be treated like export-controlled munitions, the regulatory environment for AI development has shifted fundamentally. Labs may face pressure to seek preapproval from government agencies before releasing advanced models.

Frequently Asked Questions

Why did the Trump administration force Anthropic to shut down Claude 5?

The Commerce Department issued a directive after Amazon and other tech companies claimed researchers found a way to bypass the models' safety guardrails for cybersecurity queries. The government cited national security concerns about foreign actors exploiting the vulnerability.

Did Anthropic agree with the government's assessment?

No. Anthropic said the government provided no detailed technical evidence. The company stated the alleged vulnerabilities were minor, previously known, and reproducible by other publicly available models.

What is the legal basis for shutting down an AI model?

The Commerce Department invoked national security protocols and export control regulations that bar access by foreign nationals to certain technologies. The legal framework remains unclear and is likely to face scrutiny.

How did the shutdown affect Anthropic?

Internal developer productivity reportedly dropped by 60% due to the restriction of model access for foreign nationals, many of whom work at Anthropic. Market uncertainty added to the pressure.

Could this happen to other AI companies?

Yes. The precedent suggests the government can intervene quickly against any AI lab if officials deem a model a national security risk, even over the company's objections.

Source: Fast Company / Mark Sullivan