Anthropic export controls center on SK Telecom involvement

Tom's Hardware reports that SK Telecom was listed among roughly 150 organizations added to Anthropic's internal access-control process, Project Glasswing, in early June. In a public announcement dated June 12, Anthropic said the U.S. government issued a directive to suspend access to Fable 5 and Mythos 5 by any foreign national, inside or outside the United States, and that the company disabled those two models for all customers to comply with the order. Anthropic's statement says access to its other models was not affected.

Anthropic's June 12 statement frames the government's action as tied to a demonstrated jailbreak technique the company reviewed; Anthropic wrote that the technique exposed a small number of previously known, relatively simple vulnerabilities and that other publicly available models can be bypassed in similar ways. The company said its red-team testing of Fable 5 involved thousands of hours with multiple external partners prior to launch.

The Washington Post reports the White House considered sanctions on Anthropic weeks before the directive, citing concerns after Anthropic shared access with a firm the Post describes as "suspected China-linked." Fortune reports that researchers at Amazon documented a Fable 5 jailbreak and that Amazon CEO Andy Jassy raised the finding on a scheduled call with White House officials, which helped draw Treasury and other senior officials into the matter. Anthropic's statement says the government's letter "did not provide specific details of its national security concern."

Companies operating frontier generative models that grant selective external access have previously faced heightened government scrutiny when third-party access intersects with international partners; selective-access lists, third-party red teaming, and rapid vulnerability disclosures can trigger interagency reviews when a model is perceived to carry systemic risk. Teams running comparable partner-access or enterprise-vetting programs should treat this as a signal that a single third-party jailbreak finding can escalate to binding government action within weeks, and should watch whether other providers disclose comparable access lists or change vetting procedures after this episode.

Watch for formal guidance from Treasury or the Commerce Department clarifying the criteria used to block model access, since current public accounts describe an ad hoc directive rather than a published rule. The Korea JoongAng Daily reports an Anthropic executive is confident access can be re-enabled within days; watch for that restoration and for any additional government directives or statements from Anthropic or affected partners.

Public reporting attributes the escalation to a mix of a documented jailbreak, third-party reporting, and national-security concern, but Anthropic's own statement confirms the government's letter did not spell out its specific national-security rationale, so the full basis for the directive is not yet public. The episode shows AI export-control enforcement happening ad hoc rather than through a published framework, which raises predictability concerns for any AI lab running partner-access or selective-distribution programs.