The Trump administration’s decision to impose stringent export controls on Anthropic’s most advanced artificial intelligence technology followed an escalating dispute regarding the company’s decision to grant South Korean telecommunications giant SK Telecom access to its Claude Mythos model. According to sources familiar with the internal deliberations, U.S. officials expressed significant alarm over what they characterized as SK Telecom’s historical and structural ties to the People’s Republic of China. These concerns reached a critical threshold when Amazon, a major investor and partner of Anthropic, subsequently flagged significant security vulnerabilities in Fable 5, a public-facing version of the technology. The convergence of perceived geopolitical risks and technical weaknesses ultimately prompted the White House to conclude that the AI laboratory could not be trusted to independently safeguard its dual-use technology, leading to an unprecedented order that has effectively shuttered Anthropic’s flagship models.
The Catalyst: Project Glasswing and the SK Telecom Dispute
The friction between Anthropic and the federal government originated with Project Glasswing, an initiative designed to provide a small, vetted group of organizations with early access to Claude Mythos. Mythos is widely regarded as one of the most capable AI models in existence, particularly in the realm of identifying software vulnerabilities and assisting in complex cybersecurity operations. Because of these capabilities, Anthropic initially restricted access to a handful of trusted partners.
Earlier this month, Anthropic expanded Project Glasswing to include approximately 150 companies. Among the new participants was SK Telecom, South Korea’s largest wireless carrier. Anthropic’s decision to include SK Telecom followed what the company described as weeks of collaboration with external experts and U.S. government stakeholders. Other South Korean entities, including Samsung Electronics and the Korea Internet and Security Agency (KISA), were also granted access.
However, the inclusion of SK Telecom immediately drew scrutiny from the White House. Sources close to the administration indicate that officials were concerned that SK Telecom could serve as a conduit for sensitive AI capabilities to reach China. Shortly after the expansion was announced, the White House formally requested that Anthropic revoke SK Telecom’s access to the Mythos model. While Anthropic complied immediately, the incident damaged the rapport between the AI lab and national security officials, setting the stage for more drastic regulatory intervention.
Technical Vulnerabilities: The Amazon Report
The geopolitical concerns were compounded by a technical report submitted to the White House by researchers at Amazon. Amazon, which has invested billions of dollars into Anthropic and provides the infrastructure for its models through Amazon Web Services (AWS), identified what it described as critical "jailbreak" vulnerabilities in Fable 5.
Released on June 9, Fable 5 was intended to be a highly safeguarded, public-facing iteration of the Mythos model. It featured extensive guardrails designed to prevent users from accessing the model’s more dangerous cyber-capabilities. However, Amazon’s researchers claimed they successfully circumvented these protections, allowing them to tap into the underlying power of the Mythos engine. While Anthropic and some independent cybersecurity experts argued that these types of vulnerabilities are inherent to all large language models (LLMs) and not unique to Claude, the White House viewed the report as evidence that Anthropic’s safety protocols were insufficient for a model of such high strategic value.
The Lutnick Letter and the Revocation of Access
The situation culminated on Friday when the Trump administration issued a formal order, often referred to as the "Lutnick Letter" after Secretary of Commerce Howard Lutnick, demanding that Anthropic immediately restrict access to both Claude Mythos and Fable 5. The order was remarkably broad, requiring the company to revoke access for all foreign nationals. Critically, this mandate extended not only to users abroad but also to foreign nationals and immigrants residing and working within the United States.
Faced with the prospect of implementing a nationality-based gating system—a process that would require intrusive data collection and potentially violate privacy standards—Anthropic leadership determined that the only viable path was to disable the models entirely. This decision has resulted in a significant disruption for Anthropic’s enterprise clients and research partners, and the company remains locked in tense negotiations with the White House to find a compromise that would allow the models to return to service.
Analyzing the SK Telecom-China Connection
The administration’s focus on SK Telecom highlights the complex web of corporate and geopolitical interests in the global technology sector. While SK Telecom itself has a limited direct footprint in China—reporting only $1.9 million in revenue from the country in 2024 and employing just seven people there—it is a subsidiary of SK Group, one of South Korea’s largest conglomerates. SK Group maintains extensive business interests in China, particularly in the semiconductor and energy sectors, which have long been points of contention in U.S. trade policy.
Furthermore, SK Telecom has a decades-long history of partnership with Chinese state-owned enterprises. In 2004, it formed UNISK, a joint venture with China Unicom, to provide wireless internet services. At its peak, SK Telecom held a 6.6 percent equity stake in China Unicom’s Hong Kong-listed unit. Although the partnership began to wind down in 2009 when SK Telecom sold the majority of its stake back to the Chinese carrier, it still maintains a financial interest in UNISK valued at approximately $17 million.
The U.S. government has viewed China Unicom with increasing hostility in recent years. In 2021, the first Trump administration restricted U.S. investment in the carrier, citing links to the Chinese military. More recently, in April of this year, the Federal Communications Commission (FCC) proposed a ban on U.S. telecom firms interconnecting with China Unicom, citing national security risks. From the perspective of the White House, any firm with even a residual connection to China Unicom represents a potential security leak for cutting-edge AI technology.
Chronology of the Anthropic-Government Standoff
The timeline of events illustrates a rapid deterioration of trust between the private sector and the executive branch:
- August 2023: SK Telecom invests $100 million in Anthropic, establishing a commercial partnership to develop a telecom-specific AI model.
- Early June 2024: Anthropic expands "Project Glasswing," granting SK Telecom and 149 other organizations access to the Claude Mythos model.
- June 9, 2024: Anthropic releases Fable 5, the public-facing version of Mythos, with enhanced guardrails.
- Mid-June 2024: Amazon researchers flag vulnerabilities in Fable 5 to the White House, alleging that the model’s cybersecurity safeguards can be bypassed.
- June 12-14, 2024: The White House demands Anthropic revoke SK Telecom’s access. Anthropic complies.
- June 15, 2024: Reports surface in the Washington Post and other outlets regarding the administration’s alarm over "a South Korean telecommunications company."
- June 20, 2024: The Trump administration issues a formal order (the Lutnick Letter) requiring Anthropic to bar all foreign nationals from accessing Mythos and Fable 5.
- June 21, 2024: Anthropic disables the models entirely, citing the impracticality of nationality-based restrictions.
Responses from Involved Parties
Anthropic has declined to provide official comments on the record, though sources close to the company suggest leadership is frustrated by what they view as a moving target of regulatory compliance. The company reportedly feels it acted in good faith by collaborating with the government during the Glasswing rollout, only to have the administration pivot toward more aggressive controls.
SK Telecom has pushed back against the characterization of its business. In a statement to a South Korean newspaper, the company asserted that claims of its ties to China "lack verified facts" and emphasized its status as a reliable partner to U.S. firms.
The White House has maintained a stern posture, with officials emphasizing that the protection of "frontier" AI models is a matter of national survival. While not commenting on the specific details of the Anthropic order, administration spokespeople have consistently stated that the U.S. will use all available tools, including export controls and executive orders, to prevent "adversarial nations" from gaining an edge in AI.
Broader Implications and Analysis
The shutdown of Claude Mythos and Fable 5 marks a watershed moment in the "Silicon Curtain" era of technology policy. Several key implications emerge from this development:
1. The End of Voluntary Compliance: For years, the U.S. government and AI labs operated on a basis of voluntary commitments and "red-teaming" exercises. The Anthropic case suggests that the Trump administration is moving toward a more dirigiste approach, where the government takes a direct hand in determining who can use specific models.
2. Impact on the Global AI Talent Pool: By extending the ban to foreign nationals within the U.S., the administration is creating a significant hurdle for the AI industry’s workforce. Many of the world’s leading AI researchers are immigrants or H1-B visa holders. If these individuals are legally barred from working on or even using the most advanced models, it could trigger a "brain drain" as talent moves to jurisdictions with fewer restrictions.
3. Strained Alliances: The targeting of a South Korean company—an entity from a key U.S. treaty ally—indicates that the administration’s "America First" tech policy may not grant exceptions for allies if their corporate interests overlap with Chinese markets. This could complicate diplomatic efforts to form a unified democratic front on AI governance.
4. The Precedent for Big Tech Rivalries: The role of Amazon in flagging vulnerabilities to the government adds a layer of corporate intrigue. While ostensibly a move to ensure safety, it highlights how competitive dynamics between AI providers (Amazon/Anthropic vs. Microsoft/OpenAI vs. Google) can intersect with national security reporting to influence regulatory outcomes.
As of this writing, the Claude Mythos and Fable 5 models remain offline. The outcome of the ongoing negotiations between Anthropic and the White House will likely set the precedent for how the next generation of "frontier" models is regulated, distributed, and restricted on the global stage. For now, the message from Washington is clear: in the race for AI supremacy, national security interests will take precedence over commercial expansion and international collaboration.
