SK Telecom has been pulled into a disputed fight over Anthropic Mythos export controls, after claims that the South Korean carrier’s access to Anthropic’s advanced AI security model was revoked shortly before broader restrictions affected foreign nationals.
The central allegation is that SK Telecom was included in an Anthropic vulnerability-detection program known as Project Glasswing, then had access removed after national-security concerns were raised in Washington. Those specific claims have not been independently verified, and SK Telecom has denied having ties to China.
The dispute matters because it sits at the intersection of three pressure points that are increasingly hard for AI companies to separate: frontier-model safety, cybersecurity use cases, and export-control policy. Models that can help identify software flaws are useful to defenders, but they can also raise obvious concerns if governments believe access could benefit strategic rivals.
Why SK Telecom is under scrutiny
SK Telecom is South Korea’s largest wireless carrier, but the China-related concerns described around the company appear to center less on its own direct operations and more on broader ties inside SK Group. The carrier’s direct footprint in China has been described as small, including limited revenue and staffing there, though those figures should be treated cautiously because they have not been independently verified here.
The company’s older China history is also part of the discussion. SK Telecom previously formed a wireless joint venture called UNISK with China Unicom in 2004, later invested in China Unicom convertible bonds, and eventually sold a converted stake. A remaining UNISK investment has also been described in later filings. Those details help explain why the company could attract attention in a policy fight, even if they do not by themselves establish the alleged national-security risk.
SK hynix, another SK Group company, is also relevant to the broader AI supply chain conversation. Alongside Samsung, it is one of the major Korean semiconductor companies connected to the memory and silicon ecosystem behind AI infrastructure. Both companies have been described as strategic investors in Anthropic, adding another layer to the relationship between U.S. AI labs and Korean technology firms.
The Fable 5 and Mythos 5 dispute
The SK Telecom dispute appears to be separate from a broader fight over Anthropic’s Fable 5 and Mythos 5 models. A reported demonstration involving Fable 5 allegedly showed the model being prompted to inspect a codebase and repair flaws, effectively turning it toward vulnerability discovery. That demonstration was later said to have reached U.S. officials through Amazon, Anthropic’s largest investor.
The resulting policy move was described as a Commerce Department order dated June 12 that barred foreign nationals, including immigrants inside the United States, from accessing Fable 5 and the underlying Mythos 5. Rather than attempting to sort users by nationality, Anthropic was said to have disabled access to both models more broadly.
Anthropic’s position, as described in the dispute, is that the reviewed demonstration surfaced only a small number of previously known, minor vulnerabilities. The company also argued that restricting a capability found in other frontier models could disrupt AI deployments well beyond a single product line.
That argument has support from parts of the cybersecurity community. Roughly 100 security professionals, including former Facebook security chief Alex Stamos and Luta Security founder Katie Moussouris, signed an open letter arguing that Mythos-class models may be strong at finding and weaponizing flaws but are not uniquely capable compared with other advanced models. The group called for the restrictions to be lifted.
A policy fight with business consequences
The practical problem for Anthropic is that a model built for high-end security work can become politically sensitive very quickly. Vulnerability detection is a legitimate defensive use case, especially for large companies trying to audit code at scale. But the same capabilities are difficult to fence neatly once officials begin treating access as an export-control issue.
For Korean technology companies, the dispute also shows how AI partnerships can be complicated by older corporate footprints, parent-company exposure, and semiconductor supply-chain politics. SK Telecom has reportedly rejected the allegations, saying anonymous claims in foreign media lack verified facts and that it has no ties to China.
Anthropic’s South Korea expansion adds another wrinkle. The company opened a Seoul office on June 17 and signed a memorandum of understanding with South Korea’s Ministry of Science and ICT, with SK Telecom named among local partners. That makes the access dispute more than a narrow account-permission fight: it touches Anthropic’s regional growth plans, its security-product strategy, and its relationship with U.S. policymakers.
For now, the safest reading is that SK Telecom is facing scrutiny in a fast-moving export-control controversy, while the most serious allegations around its Mythos access remain contested. The broader issue is not going away: as AI models become more capable at software analysis, governments and AI companies will keep arguing over whether those tools should be treated like security products, dual-use technology, or something in between.