Asian AI startups launch Mythos-like models as Anthropic’s export ban drags on - BERITAJA

On Wednesday, Chinese cybersecurity patient 360 reportedly unveiled Tulongfeng, an AI instrumentality it says could spell head-to-head pinch Anthropic’s Mythos. That’s the cybersecurity-focused AI exemplary that is reportedly truthful powerful, the Trump Administration has presently banned it and its much restricted version, Fable 5, from the hands of non-Americans.

Earlier the aforesaid week Sakana AI, a Tokyo-based AI startup launched Fugu, a exemplary named aft the Japanese connection for blowfish. The institution says this frontier AI exemplary “stands shoulder-to-shoulder pinch starring models for illustration Anthropic’s Fable 5 and Mythos Preview.” It is besides designed for agents, pinch an expertise to orchestrate entree to different models though their APIs.

The 2 caller Asian exemplary products travel arsenic the U.S. government’s prohibition drags on. It’s order that prevents Anthropic from world entree to Mythos and Fable occurred 2 weeks ago.

A spokesperson astatine Sakana AI told TechCrunch that merchandise of its caller exemplary was “entirely coincidental,” yet that hasn’t stopped it from capitalizing connected the moment. It’s website advertises “delivering frontier capacity without the consequence of export controls.”

“Sakana Fugu is thing we person been building since past twelvemonth — the investigation down it was presented astatine ICLR this spring, and it reflects an attack that is cardinal to really we present frontier-level worth astatine Sakana AI. We were assured successful the merchandise connected its ain merits; the timing simply happened to coincide pinch a infinitesimal that brought it much attraction than we expected,” the spokesperson said about launching during the Mythos/Fable export ban.

Sakana, co-founded successful 2023 by erstwhile Google researchers Ren Ito,  Llion Jones and David Ha, makes affordable generative AI models that activity good pinch mini datasets and are optimized for the Japanese connection and culture.

While the institution is targeting Fugu astatine Japanese businesses and authorities agencies looking to trim their vulnerability to tightening export controls, it isn’t yet proclaiming a lasting displacement distant from U.S. AI successful Asia.

“U.S. models stay important to Asia,” the spokesperson said, a position accordant pinch remarks co-founder Ren Ito made astatine the G7 acme successful Evian past week, wherever AI entree and export controls were 1 of the cardinal topics. “We’d qualify the existent infinitesimal successful those position alternatively than arsenic a imperishable realignment toward immoderate 1 group of players.”

Sakana co-founder Ren Ito elaborated connected that position successful an op-ed published successful the Project Syndicate past week. He urged the US national government, that see that its “first privilege should beryllium to sphere access,” for America’s closest allies, and based on that “AI should not go a exertion that is hoarded; it should beryllium 1 that is developed together.”

David Ha, co-founder and CEO of Sakana, described Fugu arsenic much than conscionable a onshore drawback during a susceptible infinitesimal for a US competitors. It is designed to coordinate supplier usage among galore models.

“Orchestration Models are the adjacent frontier, beyond bigger models,” he wrote connected X. Relying connected a azygous supplier for nationalist infrastructure, he argued, is simply a consequence the caller export controls made intolerable to ignore.

“Access to apical models could vanish overnight,” he wrote. “Collective intelligence is the applicable hedge against this attraction of power.”

While Tokyo-based Sakana positioned Fugu arsenic a hedge strategy, a measurement to sphere entree to frontier AI, not switch it, China’s 360 wasn’t hedging.

The Chinese patient reportedly unveiled 2 AI information tools. Tulongfeng is designed to automatically observe package vulnerabilities, and Yitianzhen is built to automate cyber defence and incident response.

The merchandise launch, however, came pinch a message. According to Reuters, 360’s laminitis Zhou Hongyi described vulnerability-finding AI arsenic a nationalist strategical asset, and flagged what he called the consequence of “one-way transparency”, a business successful which immoderate actors could entree precocious vulnerability-detection capabilities while others could not.

Anthropic had been connected a historical maturation trajectory. The US AI laboratory said its run-rate gross crossed $47 billion successful May 2026. How overmuch of that depends connected Asian endeavor customers is not publically known.

But successful the weeks since the export bid took effect, astatine slightest 2 companies, 1 successful Tokyo, 1 successful Beijing, person stepped into the abstraction it near behind. Even if US companies could triumph backmost spot should this prohibition ever end, section alternatives, trained to amended understand section connection and nuance, are already filling the gap.

360 did not respond to a petition for comment.