The Korean Telecom Giant at the Center of Anthropic’s Mythos Controversy
Original reporting by Wired
Export controls imposed by the Trump administration on Anthropic’s most powerful artificial intelligence models, Claude Mythos and Fable 5, refer to a sweeping government order restricting access to the advanced AI for foreign nationals. This significant move followed escalating concerns regarding both the models' security and their international accessibility. The immediate catalyst was a dispute over Anthropic granting South Korean telecom giant SK Telecom access to its highly capable Claude Mythos model, raising alarms among US officials. These officials expressed concern over what they alleged were SK Telecom’s historical ties to China, despite the company's denials and currently limited direct operations there. While SK Telecom itself has minimal direct operations in China today, its parent, SK Group, maintains extensive business interests in the country, and SK Telecom previously engaged in significant joint ventures with state-owned Chinese telecom operators, some of which have faced US sanctions.
Security Flaws Revealed Compounding these geopolitical anxieties, Amazon later flagged significant vulnerabilities in Fable 5, a publicly released version of Mythos, to the White House. Amazon researchers claimed they could circumvent Fable 5’s guardrails to access Mythos’ formidable cyber capabilities. This confluence of alleged Chinese ties and identified security weaknesses led the White House to conclude it could not trust Anthropic to safeguard its advanced AI. Consequently, the administration ordered Anthropic to revoke access to Mythos and Fable 5 for all foreign nationals. Rather than implement a nationality-based restriction, Anthropic opted to disable access to the models entirely, leading to an ongoing standoff with the White House.
The unprecedented standoff between Anthropic and the White House regarding Claude Mythos and Fable 5 culminates a fraught episode, revealing the escalating geopolitical sensitivities surrounding advanced artificial intelligence. Anthropic’s decision to disable its frontier models entirely for foreign access, rather than navigating complex nationality-based restrictions, vividly illustrates the profound operational challenges and the high stakes involved when national security interests intersect with private sector innovation. Driven by concerns over SK Telecom’s alleged ties to China and Amazon’s vulnerability reports, this directive marks a significant shift, positioning powerful AI models as strategic assets on par with critical hardware or defense technologies, demanding rigorous government oversight.
Broader Implications Beyond Anthropic's immediate disruptions and strained policy relations, the broader implications for the global AI ecosystem are profound and far-reaching. This intervention establishes a formidable precedent for export controls extending to AI software and model access, moving decisively beyond traditional hardware restrictions. It compels the AI community to confront the delicate balance between fostering open research, facilitating international collaboration, and safeguarding national security imperatives. Companies developing cutting-edge AI will now undoubtedly face heightened scrutiny over their global partnerships and user bases, potentially curtailing the worldwide dissemination of their technologies and influencing the speed and direction of future AI development. This incident unequivocally solidifies the view of AI as a pivotal component of national power, ensuring governments will assert significant control over its most advanced forms, thereby shaping future innovation through an increasingly security-focused lens.
Frequently asked questions
- What led the Trump administration to restrict access to Anthropic's Claude Mythos AI?
- The US government imposed export controls on Anthropic's AI due to a combination of factors. Concerns arose when South Korean telecom giant SK Telecom, alleged to have ties to China, gained access to the Claude Mythos model. This was compounded by Amazon flagging vulnerabilities in Fable 5, a version of Mythos, which reportedly allowed circumvention of its guardrails. The White House ultimately determined it could not trust Anthropic to secure its most advanced AI technology.
- Which specific Anthropic AI models were impacted by the recent US government export controls?
- The US government's export controls specifically affected Anthropic’s Claude Mythos model and Fable 5. Claude Mythos is an exceptionally skilled AI, while Fable 5 is a safeguarded public release of Mythos. Following the order to restrict access to US nationals, Anthropic opted to disable both models entirely for all users. This decision was made because implementing nationality-based access while preserving privacy would be challenging.
- What were the US government's concerns regarding SK Telecom's alleged connections to China?
- US officials were concerned about SK Telecom's alleged connections to China, particularly after the company received access to Anthropic's Claude Mythos AI. While SK Telecom denies ties to China, its parent company, SK Group, has extensive business interests there. Furthermore, SK Telecom previously engaged in a significant joint venture with state-owned China Unicom, and maintained a financial interest in that venture even after unwinding direct stakes.