US Government Restricts Access to Anthropic’s ‘Fable’ and ‘Mythos’ AI Models The shutdown of Anthropic’s Fable 5 and Mythos 5 has turned a narrow technical dispute into a global test of who controls frontier AI and on what terms.

On June 9, Anthropic released Fable 5 to the public and Mythos 5 to a restricted cybersecurity consortium, touting Fable as its most capable generally available model and Mythos as a more powerful, defense-focused system. Within days, Amazon security researchers reported they could rephrase prompts to make Fable surface code vulnerabilities, bypassing some guardrails — a technique Anthropic argued was a minor, already‑patched issue rather than a true jailbreak.

Behind the scenes on June 13–14, Amazon CEO Andy Jassy escalated the findings to senior Trump officials, who were already worried Mythos access might have reached a China‑linked group. After tense calls with Anthropic CEO Dario Amodei failed to secure a voluntary pause, the Commerce Department sent an export‑control directive on June 12 ordering Anthropic to block “any foreign national,” including its own staff, from Fable 5 and Mythos 5. To comply, Anthropic pulled both models offline worldwide within about 90 minutes, the first time U.S. export rules had ever been used to withdraw a public frontier model.

Over the following days, Anthropic flew senior researchers to Washington for crisis talks with Commerce and White House cyber officials, hoping to restore access and clarify security standards. The administration insists the move was a necessary national‑security response; officials privately describe Anthropic as reckless and hard to communicate with, and view Mythos‑class systems as potential tools for automated hacking at unprecedented scale.

Cybersecurity leaders countered with an open letter from roughly 100 experts warning that “taking the best models away from defenders” while adversaries keep similar tools “is not safety, it is sabotage,” arguing other models can already do what Fable does. Analysts note that export curbs on encryption and spyware have historically failed, and say powerful “dangerous” models are coming regardless, whether from U.S. labs, open‑source projects, or rivals in China.

Abroad, the blackout instantly cut off Indian developers and European enterprises, stranding projects mid‑stream and supercharging calls for sovereign, non‑American AI infrastructure. Leaders from Canada, Europe, and India now frame the episode as a warning that Washington can “turn off the switch” on critical AI overnight, bolstering arguments for local clouds, open‑weight models, and new regulatory regimes beyond U.S. reach.

At home, the fight has exposed a vacuum of trusted oversight. Commentators say Anthropic was “hit by export rules nobody understands,” while policymakers and labs now race to define a referee that can decide when AI is too dangerous without relying on opaque, last‑minute power plays.

Continue reading https://foxvector.com/stories/019ee377-6758-31a6-70c5-0e8a4fe65f16