Anthropic export ban: Asian startups launch Mythos-like models

Sakana AI launched Fugu and Chinese firm 360 unveiled Tulongfeng this week, stepping into capability gaps left after a U.S. export order blocked global access to Anthropic’s Mythos and its restricted variant Fable, a move that took effect two weeks earlier. Sakana positions Fugu for Japanese customers and agents, while 360 framed vulnerability-finding AI as a strategic national asset.

What did the startups actually release?

Sakana released Fugu, a frontier model aimed at Japanese businesses and government, and 360 rolled out two security-focused tools: Tulongfeng, for automatic software vulnerability discovery, and Yitianzhen, for automated cyber defense and incident response. Sakana says Fugu “stands shoulder-to-shoulder with leading models like Anthropic’s Fable 5 and Mythos Preview,” and describes the model as built for agent orchestration and coordinating access to other models through APIs. 360’s founder Zhou Hongyi framed vulnerability-finding AI as a national strategic asset and warned about the risk of “one-way transparency.”

Why did they launch now?

The launches came in the wake of a U.S. order that prevents Anthropic from global access to Mythos and Fable, issued two weeks before these products appeared. Sakana told TechCrunch the timing was "entirely coincidental," and noted its research behind Fugu was presented at ICLR this spring. Sakana also highlights a strategy focused on models tuned for the Japanese language and culture; the company was co-founded in 2023 by David Ha and Llion Jones, both Google alumni, and Ren Ito, a former executive at Mercari and Stability AI.

How do these models position themselves versus Anthropic?

Sakana frames Fugu as a hedge against sudden loss of access to foreign models, emphasizing agent orchestration rather than simply larger scale. CEO David Ha wrote on X, "Orchestration Models are the next frontier, beyond bigger models," and argued collective intelligence hedges concentration of power because "access to top models can disappear overnight." 360 presented Tulongfeng and Yitianzhen as domestic capabilities that can locate vulnerabilities and automate incident response, signalling a more assertive, security-focused posture from a Chinese vendor.

Why it matters

Anthropic had reported a run-rate revenue crossing $47 billion in May 2026, showing how commercially significant its models had become. The U.S. export order has created a concrete market opening in Asia: local alternatives trained for language and local nuance are moving to fill demand, and some vendors are explicitly selling their products as ways to reduce exposure to export controls. That shift could change procurement decisions by regional enterprises and governments that want continuity and local-language performance.

What to watch

Monitor whether Sakana’s Fugu wins contracts with Japanese firms or agencies seeking reduced export-control risk, and whether 360’s tools are adopted by Chinese enterprises or integrated into national cybersecurity programs. Also watch for any policy moves that change Anthropic’s export restrictions, since regaining access would test how sticky these local alternatives are.