Trump Administration Imposes Export Controls on Anthropic AI Models
The Trump administration imposed export controls on Anthropic’s Claude Mythos and Fable 5 AI models following concerns over SK Telecom’s ties to China and security vulnerabilities identified by Amazon. Anthropic responded by disabling access to these models for all foreign nationals to comply with US government demands, according to sources familiar with the matter.
Why did the US government restrict Anthropic’s AI models?
US officials cited concerns regarding SK Telecom’s alleged connections to China as a primary driver for the export controls, according to people familiar with the matter. This friction intensified after Amazon researchers flagged vulnerabilities in Fable 5, a public version of the Mythos model released on June 9.
Amazon researchers claimed they could bypass guardrails in Fable 5 to access the cybercapabilities of the more powerful Mythos model. While Anthropic and external cybersecurity experts argued these risks are common across AI models, the White House determined the company could not be trusted to safeguard the technology, according to a person close to the administration.
How did the SK Telecom partnership trigger these controls?
SK Telecom, South Korea’s largest wireless carrier, gained access to Mythos after weeks of collaboration with the US government and outside experts. The company has a deep financial tie to Anthropic, including a $100 million investment in 2023 to develop AI tailored for the telecommunications industry.
The conflict stems from the broader structure of SK Group. While SK Telecom lacks large operations in China, its parent conglomerate, SK Group, maintains extensive business interests in China across the energy and semiconductor sectors. The Washington Post reported that administration officials were alarmed by these links, though SK Telecom told a Korean newspaper that claims of ties to China “lack verified facts.”
What happens next for AI export controls and nationality-based access?
The administration’s order required Anthropic to revoke access for all foreign nationals, including immigrants living within the US. Anthropic chose to disable the models entirely rather than implement a nationality-based gating system, which the company viewed as difficult to manage while preserving user privacy.
This development signals a shift toward stricter national security screenings for AI partnerships. The move suggests that the US government may now scrutinize not just the direct partner, but the parent conglomerate’s global footprint. This creates a precedent where “trusted” status can be revoked based on the geopolitical interests of a partner’s affiliates.
How do the Amazon vulnerabilities contrast with Anthropic’s position?
There is a clear divide in how the security risks are framed. Amazon researchers presented the Fable 5 vulnerabilities as a direct path to “formidable cybercapabilities” within Mythos. Conversely, a person close to Anthropic stated that the company views the Amazon vulnerabilities and the SK Telecom dispute as entirely separate issues.
Furthermore, the source noted that the official US government letter demanding access restrictions to US nationals did not explicitly mention China or SK Telecom. This suggests a multi-pronged justification for the export controls—combining specific corporate concerns with broader systemic vulnerabilities.
Comparison: The Two Triggers of Export Control
| Trigger | Source of Concern | Nature of Risk |
|---|---|---|
| SK Telecom | White House / SK Group ties | Geopolitical/China influence |
| Fable 5 | Amazon Researchers | Technical/Cyber-capability leaks |
FAQ: AI Export Controls and Claude Mythos
What is Claude Mythos?
It is one of Anthropic’s most powerful AI models, specifically noted for its ability to identify software vulnerabilities.
Why was SK Telecom’s access revoked?
The Trump administration cited concerns over the company’s ties to China via its parent organization, SK Group.
What is Fable 5?
Fable 5 is a safeguarded version of the Mythos model released to the public on June 9, which Amazon researchers claimed had bypassable guardrails.
Are the models still available?
Anthropic disabled access to Mythos and Fable 5 entirely after the government demanded they be restricted to US nationals.
To stay updated on the intersection of AI governance and national security, subscribe to our industry newsletter or leave a comment below regarding how these restrictions might impact global AI development.