Anthropic's Claude Mythos disabled after SK Telecom, Amazon concerns

Anthropic has taken the drastic step of disabling access to its advanced AI model Claude Mythos and its safeguarded variant Fable 5 after the Trump administration ordered the company to revoke access for all foreign nationals, including green-card holders. The controversy began when Anthropic granted early access to Mythos—a model exceptionally skilled at identifying software vulnerabilities—to SK Telecom, South Korea's largest wireless carrier, as part of its Project Glasswing program. SK Telecom, which invested $100 million in Anthropic in 2023, was one of roughly 150 organizations given access. US officials expressed concern over SK Telecom's alleged ties to China, pointing to the company's parent conglomerate SK Group's extensive business interests in China and a historic joint venture with China Unicom. SK Telecom has denied any link to China.

Separately, Amazon researchers identified vulnerabilities in Fable 5, claiming they could bypass some guardrails to access Mythos's cybercapabilities. Although Anthropic and external experts argued the risks weren't unique, the combination of events led the White House to question Anthropic's ability to safeguard its cutting-edge AI. On Friday, the administration ordered Anthropic to restrict access to US nationals. Instead of implementing nationality-based gating—which Anthropic argued would be difficult and privacy-invasive—the company disabled access to both models entirely. Negotiations remain deadlocked. The incident highlights the growing tension between national security concerns and the global deployment of powerful AI technologies.