Asian AI Labs Launch Models as Anthropic Ban Drags On

Sakana's Fugu is built for agents. The company says it can orchestrate access to other models through their APIs, acting less like a single brain and more like a conductor pointing work at whichever model fits. Sakana, co-founded in 2023 by former Google researchers Ren Ito, Llion Jones and David Ha, has made its name on affordable generative models that work with small datasets and are tuned for the Japanese language and culture. Fugu extends that into orchestration.

China's 360 went a harder route. It reportedly announced two tools at once: Tulongfeng, designed to automatically discover software vulnerabilities, and Yitianzhen, built to automate cyber defence and incident response. One finds the holes, the other plugs them.

How the models compare on intent

How Fugu's orchestration angle works

Sakana CEO David Ha framed Fugu as more than a land grab during a vulnerable moment for a U.S. competitor. The pitch is architectural. Rather than chasing an ever-bigger single model, Fugu is designed to coordinate agent usage among many models at once.

Ha calls orchestration the next frontier, beyond bigger models, and his reasoning is shaped directly by the ban. Relying on a single provider for national infrastructure, he argued on X, is a risk the export controls made impossible to ignore.

Access to top models can disappear overnight. Collective intelligence is the practical hedge against this concentration of power, Ha wrote.

That is the mechanical idea behind Fugu: if any one provider can be switched off by a government order, a layer that spreads work across several models becomes insurance rather than a luxury. A Sakana spokesperson stressed that U.S. models remain important to Asia, calling the release entirely coincidental even as the company's own website advertises frontier capability without the risk of export controls.

Why 360 is not hedging at all

Where Sakana frames Fugu as a way to preserve access to frontier AI rather than abandon it, 360 made no such gesture. Its launch came wrapped in a message.

According to Reuters, 360 founder Zhou Hongyi described vulnerability-finding AI as a national strategic asset and flagged what he called the risk of one-way transparency, a situation in which some actors could access advanced vulnerability-detection capabilities while others could not. Read plainly, that is an argument for building a domestic Mythos rival rather than waiting for one to be licensed back in.

For anyone running servers or shipping software, the Tulongfeng angle is the one to watch. An AI that automatically finds vulnerabilities is dual-use by nature: the same engine that helps a defender patch faster can help an attacker find an opening first. Whoever holds the better automated bug-hunter holds real leverage.

What happens over the next 24 to 72 hours

Expect the framing war to escalate before the technical claims get tested.

• Benchmarks under scrutiny. Both Fugu and Tulongfeng claim parity with banned Anthropic models. Independent evaluators will start probing those claims quickly, and head-to-head numbers, when they appear, will matter more than launch-day language.
• Policy noise. AI access and export controls were already central topics at last week's G7 summit in Evian, where Sakana co-founder Ren Ito argued the U.S. should prioritise preserving access for its closest allies. Fresh Asian launches give that debate new fuel.
• Enterprise hedging. Sakana is targeting Japanese businesses and government agencies looking to reduce export-control exposure. Watch for other regional buyers to start formal evaluations of non-U.S. options as a contingency, even if U.S. models stay in their stack for now.

The deeper question is whether this is a permanent realignment or a temporary scramble. Sakana itself is careful here, declining to declare a lasting shift away from U.S. AI in Asia and describing the moment as a stress test rather than a changing of the guard. The honest read: a two-week-old ban has already produced two frontier launches and a sharper argument that depending on any single provider is a strategic risk. Whether the models live up to the marketing is the part still to be proven.

Source: TechCrunch