US Government Orders Anthropic to Halt Access to Fable 5 and Mythos 5 AI Models The clash over Anthropic’s most powerful AI models has morphed from a technical dispute into a global test of who controls frontier AI, and on what terms.

On June 9, Anthropic released Fable 5, a safeguarded public version of its Mythos‑class cybersecurity model, while keeping the more powerful Mythos 5 restricted for vetted defenders under Project Glasswing. The model quickly topped public benchmarks and “crushed OpenAI’s GPT 5.5 on coding” tasks, making it the strongest widely available AI system.

Concerns began building even before launch. IMF chief Kristalina Georgieva warned that models like Mythos “can be used to destroy the financial system” if misused, while Anthropic’s own CEO had described Mythos as a “serious civilizational challenge.”

On June 12, Amazon security researchers demonstrated a way to prompt Fable 5 to surface code vulnerabilities using a simple “fix this code” reframing, a technique critics stressed “is not a jailbreak” and is possible in other leading models. Amazon CEO Andy Jassy escalated the findings to senior Trump administration officials, feeding into White House worries about hacking risks and even unconfirmed suspicions that China‑linked actors might access or distill Mythos.

That evening, Commerce Secretary Howard Lutnick sent Anthropic an export‑control directive barring use of Fable 5 and Mythos 5 by “any foreign national,” including Anthropic’s own staff. Because it could not practically separate users, Anthropic disabled both models worldwide, saying the letter “did not provide specific details” and that only a “narrow, non‑universal jailbreak” had been shown, with comparable behavior in other public systems.

From there, the fallout broadened fast. Cybersecurity leaders organized an open letter arguing the move “has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.” Axios reported fears that the precedent could discourage American firms from building tools that help patch vulnerabilities at scale.

Abroad, allies and rivals drew their own lessons. Indian officials and founders said the shutdown showed “what happens when your AI infrastructure runs on someone else’s politics,” fueling calls for sovereign and open‑source alternatives. European startup Mistral used the episode to bolster its pitch for AI sovereignty, warning governments against relying on US providers who “have the keys” and “can turn it off or turn it on.” Canada’s Prime Minister Mark Carney compared the concentrated reliance on a few frontier models to pre‑2008 financial “model risk” and urged diversification of AI infrastructure.

Inside the US, critics blasted the administration’s opaque process as “capricious” and potentially self‑defeating for its own AI export strategy. TechCrunch argued the ban “was never about an AI jailbreak” so much as a reactionary and possibly retaliatory use of an obscure export tool, while Axios said the saga exposes AI’s “missing referee” — no clear, trusted arbiter to judge when a model is too dangerous.

Anthropic, for its part, insists it supports statutory powers to block genuinely unsafe deployments but says this action “does not adhere” to principles of transparency and technical grounding. Company executives have flown to Washington for crisis talks with Commerce and White House staff, even as business customers, suddenly cut off mid‑project, are absorbing a different lesson: never bet everything on a single, remotely controlled AI model.

As US and European officials now discuss a “trusted partner” scheme for controlled access to future systems, the Fable–Mythos episode has turned an abstract debate over AI safety into a live demonstration of state power over code — and a catalyst for a more fragmented, sovereign AI world.

Continue reading https://foxvector.com/stories/019ed73b-f27f-3586-706b-1a1146ec202c