The U.S. government ordered Anthropic to suspend access to its new Mythos and Fable 5 AI models for foreign nationals, citing national security risks from a "jailbreak" vulnerability flagged by Amazon and concerns over a South Korean partner. Anthropic complied by shutting down the models for all users, but about 200 early testers retain access. The move, based on export control laws, has been called legally questionable by experts.
Mythos Fable shutdown directive
- ▪On June 12, 2026, the U.S. government ordered Anthropic to suspend access to its Fable 5 and Mythos 5 models for all foreign nationals, citing national security
- ▪The European Union's cybersecurity agency, ENISA, which had been invited to join Project Glasswing, was denied access following the U.S. government's order
- ▪Despite the shutdown, about 200 organizations in Anthropic's "Project Glasswing" program retained access to a preview version of Mythos
- ▪Anthropic complied by shutting down access for all users worldwide, stating it could not immediately separate affected foreign nationals from its user base
- ▪Technology firms Dragos and Cisco Systems confirmed they retained access to the Mythos Preview as part of Project Glasswing
- ▪Senior Anthropic staff met with Commerce officials and the National Cyber Director’s office in Washington, but the meeting ended without a resolution
- ▪The directive was issued just three days after Anthropic released the Fable 5 and Mythos 5 models on June 9, 2026
Amazon security vulnerability disclosure
- ▪SK Telecom denied having ties to China, stating the allegations from anonymous insiders lacked verified facts
- ▪SK Telecom previously held an equity stake in China Unicom, a state-owned Chinese operator in which the U.S. later restricted investment
- ▪U.S. officials were also concerned about Anthropic granting Mythos access to South Korean telecom company SK Telecom due to its alleged ties to China
- ▪The government's directive was prompted by a "narrow jailbreak vulnerability" and concerns the models could be used to find software weaknesses
- ▪Amazon researchers reportedly demonstrated to the White House that Fable 5 could be prompted to provide information useful for cyberattacks
Anthropic AI safety advocacy
- ▪Anthropic CEO Dario Amodei wrote that frontier AI models should be required to go through technical testing and auditing before release
- ▪Anthropic publicly disagreed with the directive, calling it a "misunderstanding" and arguing it was not transparent, fair, or grounded in technical facts
- ▪Anthropic has publicly advocated for AI regulation, including giving the government power to block models deemed unsafe
Trump administration export controls
- ▪The government's order is based on the Export Control Reform Act of 2018, which regulates the transfer of sensitive technologies to foreign parties
- ▪A former Commerce Department official noted that previous agency guidance concluded that providing foreign nationals with cloud computing access was not an export
- ▪Legal experts and former Commerce Department officials questioned if the Export Control Reform Act applies to remote access of a cloud-based AI model
Industry expert shutdown responses
- ▪Alex Stamos, former Facebook security chief, organized an open letter signed by over 150 executives calling for the directive to be lifted
Department of Defense blacklist
- ▪Anthropic is in an ongoing legal dispute with the Department of Defense over a March 2026 "supply chain risk" designation
Story comments
Loading comments…