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

On Wednesday, Chinese language cybersecurity agency 360 reportedly unveiled Tulongfeng, an AI software it says can go head-to-head with Anthropic’s Mythos. That’s the cybersecurity-focused AI mannequin that’s reportedly so highly effective, the Trump Administration has at the moment banned it and its extra restricted model, Fable 5, from the palms of non-People.

Earlier the identical week Sakana AI, a Tokyo-based AI startup launched Fugu, a mannequin named after the Japanese phrase for blowfish. The corporate says this frontier AI mannequin “stands shoulder-to-shoulder with main fashions like Anthropic’s Fable 5 and Mythos Preview.” It’s also designed for brokers, with a capability to orchestrate entry to different fashions although their APIs.

The 2 new Asian mannequin merchandise come because the U.S. authorities’s ban drags on. It’s order that forestalls Anthropic from world entry to Mythos and Fable occurred two weeks in the past.

A spokesperson at Sakana AI advised Trendster that launch of its new mannequin was “solely coincidental,” but that hasn’t stopped it from capitalizing on the second. It’s web site advertises “delivering frontier functionality with out the danger of export controls.”

“Sakana Fugu is one thing now we have been constructing since final yr — the analysis behind it was introduced at ICLR this spring, and it displays an strategy that’s central to how we ship frontier-level worth at Sakana AI. We had been assured within the product by itself deserves; the timing merely occurred to coincide with a second that introduced it extra consideration than we anticipated,” the spokesperson mentioned about launching in the course of the Mythos/Fable export ban.

Sakana, co-founded in 2023 by former Google researchers Ren Ito,  Llion Jones and David Ha, makes inexpensive generative AI fashions that work properly with small datasets and are optimized for the Japanese language and tradition.

Whereas the corporate is concentrating on Fugu at Japanese companies and authorities companies seeking to scale back their publicity to tightening export controls, it isn’t but proclaiming a long-lasting shift away from U.S. AI in Asia.

“U.S. fashions stay vital to Asia,” the spokesperson mentioned, a view in line with remarks co-founder Ren Ito made on the G7 summit in Evian final week, the place AI entry and export controls had been one of many central matters. “We’d characterize the present second in these phrases somewhat than as a everlasting realignment towards anyone set of gamers.”

Sakana co-founder Ren Ito elaborated on that view in an op-ed revealed within the Mission Syndicate final week. He urged the US federal authorities, that think about that its “first precedence must be to protect entry,” for America’s closest allies, and argued that “AI mustn’t develop into a know-how that’s hoarded; it must be one that’s developed collectively.”

David Ha, co-founder and CEO of Sakana, described Fugu as greater than only a land seize throughout a susceptible second for a US opponents. It’s designed to coordinate agent utilization amongst many fashions.

“Orchestration Fashions are the subsequent frontier, past greater fashions,” he wrote on X. Counting on a single supplier for nationwide infrastructure, he argued, is a danger the current export controls made not possible to disregard.

“Entry to high fashions can disappear in a single day,” he wrote. “Collective intelligence is the sensible hedge in opposition to this focus of energy.”

Whereas Tokyo-based Sakana positioned Fugu as a hedge technique, a option to protect entry to frontier AI, not substitute it, China’s 360 wasn’t hedging.

The Chinese language agency reportedly unveiled two AI safety instruments. Tulongfeng is designed to mechanically uncover software program vulnerabilities, and Yitianzhen is constructed to automate cyber defence and incident response.

The product launch, nevertheless, got here with a message. In response to Reuters, 360’s founder Zhou Hongyi described vulnerability-finding AI as a nationwide strategic asset, and flagged what he referred to as the danger of “one-way transparency”, a scenario through which some actors may entry superior vulnerability-detection capabilities whereas others couldn’t.

Anthropic had been on a historic progress trajectory. The US AI lab mentioned its run-rate income crossed $47 billion in Might 2026. How a lot of that will depend on Asian enterprise prospects shouldn’t be publicly identified.

However within the weeks because the export order took impact, not less than two firms, one in Tokyo, one in Beijing, have stepped into the area it left behind. Even when US firms may win again belief ought to this ban ever finish, native alternate options, educated to higher perceive native language and nuance, are already filling the hole.

360 didn’t reply to a request for remark.