The U.S. Department of Commerce’s recent export controls on Anthropic’s advanced AI model, Mythos, have unexpectedly thrust SK Telecom into the geopolitical spotlight. On the surface, this appears to be a routine case of restricting sensitive AI technology; in reality, it exposes the fragility of trust between the United States and South Korea in the semiconductor and AI supply chains—even though SK Telecom itself has not directly exported restricted technology to China. The mere historical and structural ties between its parent conglomerate, SK Group, and Chinese sectors in energy, materials, and memory chips have triggered Washington’s “associational red flags.”
This incident is not isolated. Since 2023, Uclidean export controls have expanded from chipmaking equipment to AI models themselves. Mythos, Anthropic’s latest large language model, is reportedly approaching GPT-5-level capabilities in reasoning efficiency and multimodal integration, leading U.S. officials to classify it as a “potential dual-use asset.” SK Telecom, as Mythos’ early deployment partner in South Korea, aimed to accelerate enterprise AI adoption locally—but its affiliation with SK Group’s global footprint drew scrutiny.
The core tension lies in SK Group’s complex web of operations. While SK Telecom operates independently, its sister company SK hynix is the world’s second-largest DRAM supplier outside of Taiwan, China, and maintains a significant packaging and testing facility in Wuxi, China. Although SK hynix has reduced its China-based capacity to under 15% of total output by 2024 (per its annual report) and adheres strictly to export compliance, U.S. policymakers remain concerned about potential indirect leakage of advanced memory tech to Chinese AI training clusters via third parties. Compounding matters, SK Telecom’s 2019 joint venture with China Unicom—though fully exited by 2022—still appears in risk assessments as a “legacy linkage.”
This “guilt-by-association” logic is redefining the rules of global tech collaboration. In the past, legal firewalls like separate subsidiaries could mitigate geopolitical exposure. Now, Washington is evaluating entire corporate ecosystems based on group-wide affiliations. Even if SK Telecom’s AI business is operationally siloed, its shared ownership with SK hynix renders it a “systemic node of concern.”
I judge this marks a fundamental shift in AI-era export control paradigms: from *item-based* restrictions to *ecosystem-based* containment. The target is no longer just chips or code—it’s the network of actors embedded in a technology’s value chain. Anthropic’s partnership with SK Telecom was intended to leverage Korea’s 5G infrastructure and cloud platforms for rapid model deployment. Yet U.S. regulators interpreted it as a potential conduit for technology diversion. This misalignment reveals a deeper strategic divergence between Seoul and Washington on AI governance.
Seoul’s response has been cautious. South Korea relies heavily on U.S. suppliers for semiconductor equipment and EDA tools, yet China still accounts for nearly 30% of SK hynix’s revenue. In 2025, Korean exports of semiconductor equipment components to the U.S. grew by 27%, while memory chip exports to China remained at $18 billion (Korea International Trade Association data). This “hedging strategy” is becoming untenable as AI geopolitics intensify.
More alarmingly, similar scrutiny could soon extend to other Korean firms. Samsung Electronics, though uninvolved in this case, operates a major NAND fab in Xi’an, China—a facility that may face heightened review under the same logic. If the U.S. adopts a blanket policy that disqualifies any firm with *any* commercial ties to China from participating in cutting-edge AI partnerships, South Korea’s tech sector will be forced into an all-or-nothing choice between Washington and Beijing—a strategic dilemma Seoul has long sought to avoid.
Meta (formerly Facebook), an early investor in Anthropic, has remained silent—but its position is telling. As Meta pushes global deployment of its Llama series, it too risks being ensnared by overbroad export controls. This likely explains why some Silicon Valley firms privately question the current regime’s “mission creep.”
In the long run, the SK Telecom episode underscores a harsh truth: in the global AI race, technical superiority no longer guarantees market access. Trust—specifically, geopolitical “trustworthiness”—is becoming a scarcer resource than compute power. For South Korea, rebuilding U.S. confidence in its technological neutrality without sacrificing its China market will be the defining industrial diplomacy challenge of the next five years.
The critical question remains: as AI models themselves become national security review subjects, can global technological collaboration retain even minimal openness? Or are we sliding toward a fragmented AI world divided not by technical standards, but by lines of geopolitical trust?