You cannot restrict a capability without creating a market for the workaround.
That does not mean the restriction is stupid. Sometimes restrictions are the price of not handing dangerous tools to everyone with an internet connection.
But markets are very good at reading fear.
TechCrunch reports that while Anthropic’s Fable 5 and Mythos 5 access fight was still unfolding, two Asian AI launches moved straight into the gap. Chinese cybersecurity firm 360 unveiled Tulongfeng, an AI tool it says can rival Anthropic’s Mythos. Sakana AI, from Tokyo, launched Fugu, a multi-agent orchestration system marketed as frontier capability without single-vendor dependency.
Sakana says the timing was coincidental.
Maybe it was.
The marketing was not.
Fugu’s page is explicit: it coordinates a pool of models through one API, can opt specific providers out for compliance or privacy reasons, and claims strong performance on engineering, science, and reasoning benchmarks while avoiding export-control risk. Whether every benchmark claim survives independent pressure is a separate question.
The positioning is the story.
Access risk became a feature
Until very recently, the strongest pitch in AI was raw capability. Bigger model, better benchmark, deeper reasoning, longer context, stronger coding.
Now another pitch is becoming just as important: can you keep using it tomorrow?
That is why Fugu is interesting even if you are skeptical of the benchmarks. It is not just saying “we are smarter.” It is saying “we route around dependency.” That is a different product category. It turns orchestration, provider choice, regional availability, and compliance flexibility into part of the value proposition.
This is exactly what happens when frontier models become conditional infrastructure. The moment access feels fragile, customers start valuing boring things again: portability, resilience, fallback paths, and control.
The irony is sharp
Export controls are meant to preserve advantage and reduce risk.
But they can also advertise the size of the prize. If a model is so powerful that it gets pulled from global access, every serious competitor receives a free market map. Build something near that capability. Package it for the customers who are nervous about US access rules. Talk about sovereignty, routing, and independence. Boom, you have a wedge.
This does not mean Sakana or 360 have truly matched Mythos. Claims are cheap, and cyber capability claims are especially slippery.
But it does mean the narrative has shifted.
The frontier is no longer one clean ladder where everyone waits for the American lab to release the next rung. It is becoming a messy system of gated models, orchestrators, regional alternatives, specialized tools, and buyers who want less single-provider exposure.
That is less elegant.
It is probably more real.
Sources: TechCrunch, Sakana AI, Quartz