Anthropic crisis: SK Telecom China ties and Fable 5 flaws
The White House ordered Anthropic to cut SK Telecom’s access to Claude Mythos over alleged China ties; reported Fable 5 safety flaws then.
TL;DR
- 01The White House ordered Anthropic to cut SK Telecom’s access to Claude Mythos over alleged China ties; reported Fable 5 safety flaws then.
- 02Anthropic shut down its Claude Mythos and Fable 5 models after US officials raised alarm about SK Telecom’s alleged ties to China and security flaws in Fable 5.
- 03The White House told Anthropic to cut off SK Telecom’s access to Mythos, Anthropic complied immediately, and subsequent security reports prompted both models to be taken completely offline.
Anthropic shut down its Claude Mythos and Fable 5 models after US officials raised alarm about SK Telecom’s alleged ties to China and security flaws in Fable 5. The White House told Anthropic to cut off SK Telecom’s access to Mythos, Anthropic complied immediately, and subsequent security reports prompted both models to be taken completely offline.
How did SK Telecom get access to Anthropic’s models?
SK Telecom had access to Claude Mythos through Anthropic’s partner program Project Glasswing. US officials grew alarmed over what they saw as SK Telecom’s alleged ties to China, and the White House told Anthropic to cut off SK Telecom’s access; Anthropic complied right away. SK Telecom denied any China connections to a Korean newspaper.
Project Glasswing is named in the reporting as the channel through which the access occurred. The broader corporate context noted in the reporting: SK Telecom is part of SK Group, which has major business interests in China and held a stake in the state-owned Chinese carrier China Unicom until 2009.
What else pushed Anthropic to take models offline?
Amazon and other companies flagged security flaws in Fable 5 that could be used to bypass safety restrictions. Those reported vulnerabilities came on top of the national-security concerns tied to SK Telecom, and together the incidents caused the White House to lose confidence in Anthropic and force both models completely offline.
The reporting links the two threads directly: the China-ties concern prompted the White House to demand access be cut, and the subsequent discovery of bypassable safety flaws in Fable 5 — raised by Amazon and other unnamed companies — intensified officials’ loss of confidence.
Why does this matter?
The episode ties two different risk classes into a single outcome: geopolitical concern about a partner’s business links, and technical security flaws in a deployed model. That combination persuaded the White House to intervene directly and remove access and availability, showing that both supply-chain relationships and model safety reports can sway national-level decisions.
For companies building and partnering on large models, the immediate lesson is operational: partner vetting and rapid, credible remediation of flagged vulnerabilities matter not just to product safety but to continuity of access and market standing.
What to watch
Watch for any public remediation from Anthropic addressing the Fable 5 flaws flagged by Amazon and others, and for clarifications from SK Telecom about its corporate ties following its denial to a Korean newspaper. A White House decision to restore confidence would be the clearest signal that officials’ concerns have been satisfied.
Written by The Brieftide · Source: The Decoder
The Brieftide Daily · 06:00
Briefs like this one, in your inbox every morning.
Continue reading
More in AI SafetyAI4SE and SE4AI: A decade review of AI in systems engineering
H. Sinan Bank, Daniel R. Herber and Thomas Bradley map three research phases and assess 1.
Deepmind AI Control Roadmap: agents treated as insider threats
Deepmind ties permissions to verified behavior, models agents as rogue employees.
Dario Amodei's AI playbook: Anthropic's regulation plan
Amodei urges binding third-party audits, federal power to block risky models, export controls.
Germany approves DE-AISI, an AI security institute based on UK
The National Security Council authorised a German AI Security Institute to test advanced models.