Sakana and 360 launch Mythos-style AI as US export ban bites

This story crystallizes how quickly AI power vacuums get filled. When the U.S. government forced Anthropic to pull Mythos and Fable 5 for foreign users, it didn’t just dampen capability – it created a market opening. Within days, a major Chinese cybersecurity firm and a Japanese research‑heavy startup stepped in with tools explicitly marketed as Mythos‑class alternatives, aimed squarely at governments and enterprises worried about getting locked out of U.S. models.

Sakana’s Fugu is especially interesting from an AGI‑race perspective. It’s not just another big model; it’s an orchestration layer that routes work across a pool of agents, including its own systems, to deliver frontier‑level performance while remaining resilient to export controls. That architecture – an agentic meta‑model coordinating many specialized models – is one plausible shape of post‑LLM AI systems.

Strategically, these launches underscore that export controls don’t stop capability; they redistribute it. Asia’s message is that it won’t wait around while Washington decides who is “trusted” enough for U.S. frontier models. For U.S. labs and policymakers, the risk is obvious: push too hard on access restrictions and you accelerate the build‑out of rival ecosystems that are less inclined to align with Western safety and governance norms.