Korean Media: White House Discovers South Korean Telecom Company Linked to China on Anthropic’s Latest Model Access List… Permission Revoked
¬ Erosion of Trust May Become Decisive Trigger for Export Controls
Recently, the U.S. government issued a mandatory regulation prohibiting foreign nationals from accessing Anthropic’s top-tier high-performance AI models, "Mythos5" and "Fable5." On the 15th, according to The Washington Post (WP), a suspected South Korean telecommunications company with alleged ties to China was identified on a list of access authorizations submitted by Anthropic—prompting consideration of sanctions. The Trump administration's move to impose export controls on Anthropic’s latest models stems from ongoing disagreements over "AI militarization" since Trump’s second term, as well as deteriorating trust in Anthropic by the White House.
According to WP reports, Anthropic submitted a list of 111 institutions that were granted priority access to the latest model "Mythos5" prior to the Trump administration’s sanctions. As Anthropic expanded the list, an additional 50 organizations were granted access. From this point onward, senior officials within the Trump administration began discussing measures to reclaim the technology via export controls. "Trust in Anthropic at the White House has been rapidly eroding," WP cited government sources as saying. The decisive turning point came when a South Korean telecom company allegedly linked to China appeared on Anthropic’s submitted access list.
In May, Anthropic expanded access to Mythos across more countries and granted broader permissions. The model’s security vulnerability detection capability rivals that of human experts. The “Glass Wing Initiative,” a security protocol built upon Mythos, added 150 new partner institutions from 15 countries—including three South Korean companies: Samsung Electronics, SK Hynix, and SK Telecom. Additional South Korean media reported that the Korea Internet & Security Agency (KISA) also participated. Although WP did not explicitly name the "suspected South Korean telecom company with links to China," an anonymous source closely associated with Anthropic told WP: "The company’s Mythos access privileges were immediately revoked," adding, "They [Anthropic] excessively broadened the scope of access."
On the 11th, Amazon CEO Andy Jassy raised concerns about “jailbreaking” to the Trump administration—an issue that ultimately served as the decisive trigger for the U.S. government’s stringent export control measures against Anthropic’s AI models. Anthropic CEO Dario Amodei attempted to clarify misunderstandings with senior Trump administration officials, but during a meeting chaired by Shawn Kaincross, Director of the White House Office of National Cybersecurity (ONCD), export controls were adopted as the most direct response. Former White House AI advisor David Sachs stated: "The government asked Amodei to fix the vulnerabilities and recall the model, but he refused." Politico reported: "After NSA assessments, White House officials increasingly felt mounting evidence validated Jassy’s concerns."
Source: Chosun Ilbo
Original: toutiao.com/article/1868219175539722/
Disclaimer: This article reflects the personal views of the author