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

The recent unveiling of Tulongfeng by China's 360 and Fugu by Japan's Sakana AI marks a significant development in the AI landscape, particularly in the wake of the US export ban on Anthropic's Mythos and Fable 5. Specifically, 360's Tulongfeng is designed to automatically discover software vulnerabilities, while Sakana AI's Fugu is positioned as a frontier AI model capable of orchestrating access to other models through their APIs. Notably, Fugu's launch coincides with the US export ban, with Sakana AI's spokesperson describing the timing as "entirely coincidental." However, the company is capitalizing on the moment, advertising its ability to deliver frontier capability without the risk of export controls. Sakana AI's co-founders, Ren Ito, Llion Jones, and David Ha, have emphasized the importance of preserving access to frontier AI, with Ito urging the US government to consider the need to preserve access for its closest allies.

The emergence of these Asian AI models reflects a broader trend of regional players seeking to reduce their dependence on US AI technology, driven in part by the tightening of export controls. This shift is evident in Sakana AI's focus on developing affordable generative AI models optimized for the Japanese language and culture, as well as 360's emphasis on vulnerability-finding AI as a national strategic asset. The export ban on Anthropic's models, which had achieved a historic growth trajectory with a run-rate revenue of $47 billion, has created an opportunity for local alternatives to fill the gap. With Anthropic's models no longer accessible to non-US entities, Asian startups are positioning themselves to capitalize on the demand for advanced AI capabilities.

The implications of this development are significant, with potential risks and opportunities arising from the increasing regionalization of AI. As local alternatives, trained to better understand local language and nuance, continue to fill the gap left by Anthropic's models, US companies may face challenges in winning back trust should the export ban be lifted. Furthermore, the emphasis on orchestration models, as highlighted by Sakana AI's Fugu, suggests a shift towards more coordinated and collective approaches to AI development. As David Ha noted, relying on a single provider for national infrastructure poses significant risks, and the recent export controls have made it impossible to ignore the need for a more diversified approach.