Asian AI Startups Unveil Mythos-Inspired Models Amid Ongoing Export Restrictions on Anthropic

On Wednesday, the Chinese cybersecurity company 360 announced the launch of Tulongfeng, an AI tool it claims can rival Anthropic’s Mythos. This cybersecurity-focused AI model has been deemed so powerful that the Trump Administration has banned it, along with its more restricted version, Fable 5, from non-American access.

Earlier in the week, Sakana AI, a startup based in Tokyo, introduced Fugu, an AI model named after the Japanese term for blowfish. The company asserts that this advanced model “stands shoulder to shoulder” with leading offerings like Anthropic’s Fable 5 and Mythos Preview. It is intended for agents, providing the capability to manage access to other models through their APIs.

These new AI models from Asia arise as the U.S. government’s export ban remains in effect. The restrictions that prevent Anthropic from global access to both Mythos and Fable were enacted two weeks ago.

A representative from Sakana AI stated that the release of Fugu was purely coincidental, but the company has nonetheless seized the opportunity to promote its offerings. Their website advertises that they provide advanced capabilities without the risks associated with export regulations.

“Sakana Fugu is something we have been developing since last year—research related to it was presented at ICLR this spring, and it represents our commitment to delivering cutting-edge value at Sakana AI. We were confident in the product’s benefits; the timing simply aligned with a moment that garnered us more attention,” said the spokesperson regarding the launch during the export ban affecting Anthropic.

Founded in 2023 by former Google researchers Ren Ito, Llion Jones, and David Ha, Sakana aims to offer cost-effective generative AI models optimized for smaller datasets and designed to cater to Japanese language and culture.

While the company targets Fugu towards Japanese enterprises and government entities seeking to mitigate the impact of tightening export regulations, they have not declared a permanent shift away from U.S. AI in Asia.

“U.S. models continue to hold significance for Asia,” the spokesperson asserted, aligning with remarks made by co-founder Ren Ito at the recent G7 summit in Evian, where access to AI and export restrictions were major discussion points. “We believe it’s more accurate to view the current situation as a momentary development rather than a lasting realignment towards any specific group of providers.”

In an op-ed published last week, Ren Ito emphasized the need for the U.S. government to prioritize maintaining access for its closest allies and warned against treating AI as a resource to be hoarded. He argued for collaborative development in the field.

David Ha, co-founder and CEO of Sakana, described Fugu as more than a strategic move to capitalize on competitors’ vulnerabilities. It is engineered to coordinate the use of various models by different agents.

“Orchestration Models present the next frontier, beyond large models,” he mentioned on social media. He cautioned that depending solely on one provider for national infrastructure has become untenable due to recent export controls.

“Access to leading models can vanish overnight,” he warned. “Collective intelligence serves as a practical safeguard against this concentration of power.”

While Sakana positioned Fugu as a strategy to preserve access to advanced AI, 360’s intentions were more aggressive.

The Chinese firm introduced two cybersecurity tools: Tulongfeng for automatically identifying software vulnerabilities and Yitianzhen for automating cyber defense and incident response.

Zhou Hongyi, the founder of 360, stated that vulnerability-finding AI is a national strategic asset and highlighted the dangers of “one-way transparency,” where certain entities may access advanced vulnerability detection while others cannot.

Anthropic had seen rapid growth, achieving a projected revenue run rate of $47 billion by May 2026. However, the proportion of this revenue originating from Asian enterprise clients remains unclear.

In the weeks since the export restrictions took effect, both a Tokyo-based and a Beijing-based company have ventured into the space left vacant by Anthropic. Even if U.S. firms manage to regain trust if the ban ends, local alternatives with training tailored to regional languages and nuances are already filling the void.

360 declined to comment when approached for a statement.