Asian AI Startups Launch Mythos Rivals Amid Anthropic Export Ban
Chinese and Japanese firms move to capture the frontier AI market as US export controls block Anthropic's most powerful models.
Asian AI startups are aggressively filling the void left by U.S. export controls, launching "frontier-class" models designed to rival Anthropic’s banned Mythos series. This week, China’s 360 and Japan’s Sakana AI unveiled powerful new tools that promise high-end reasoning and cybersecurity capabilities without the risk of American regulatory interference. This shift matters because it signals a permanent fragmentation of the global AI market, where U.S. labs are losing their dominance in Asia to local players who prioritize regional sovereignty and uninterrupted access for enterprise clients.
Key Details
The rapid emergence of these rivals comes just two weeks after the U.S. government implemented a strict export ban on Anthropic’s most advanced models, Mythos and Fable 5. These models, which had become industry standards for high-level reasoning and autonomous agents, are now legally inaccessible to non-American entities, creating a massive vacuum in the Asian enterprise sector.
Chinese cybersecurity giant 360 was the first to strike, unveiling "Tulongfeng," an AI tool specifically designed to automate the discovery of software vulnerabilities—a core capability of the banned Mythos model. Alongside it, 360 introduced "Yitianzhen," a platform for automated cyber defense and incident response. 360 founder Zhou Hongyi has characterized these tools as "national strategic assets," emphasizing the need to avoid "one-way transparency" where only the West possesses advanced AI security capabilities.
Simultaneously, Tokyo-based Sakana AI launched "Fugu," a model named after the Japanese blowfish. Fugu is positioned as a direct competitor to Anthropic’s Fable 5, with a specific focus on agent orchestration. Unlike static models, Fugu is designed to coordinate multiple AI systems through their APIs, allowing businesses to build complex autonomous workflows that are immune to sudden changes in U.S. export policy.
What This Means
This isn't just a temporary workaround; it is a fundamental realignment of the AI landscape. For years, the global tech industry operated on the assumption that U.S. frontier models would be the universal substrate for the next generation of software. The recent export controls have shattered that trust.
Asian companies and governments now view reliance on a single, politically volatile provider as an unacceptable business risk. By launching local alternatives, startups like Sakana and 360 are offering more than just compute; they are offering "regulatory insurance." Even if the U.S. eventually lifts the ban, the trust has been broken, and the local ecosystems being built today—optimized for local languages and cultural nuances—will be difficult for American labs to displace.
Technical Breakdown
The new models from Asia represent a shift from "bigger is better" to "smarter orchestration."
- Orchestration Layers: Sakana’s Fugu model prioritizes the ability to manage other AI agents, treating the model as a "manager" rather than just a "worker."
- Vulnerability Detection: 360’s Tulongfeng utilizes reinforcement learning to simulate attack vectors and identify code flaws with a precision previously only seen in closed-source U.S. military-grade AI.
- Data Efficiency: Sakana AI, founded by former Google researchers, uses specialized training techniques that allow frontier-level performance on smaller, more focused datasets, making their models cheaper to run and easier to customize for Japanese enterprise needs.
- Sovereign Infrastructure: Both firms are building their stacks to run on local hardware, reducing dependency on U.S.-based cloud providers like AWS or Azure.
Industry Impact
The financial stakes are astronomical. Before the ban, Anthropic was on a trajectory to hit a $47 billion revenue run-rate by the end of 2026. A significant portion of that growth was expected to come from Asian markets. Now, that revenue is being diverted to local players.
Furthermore, this fragmentation accelerates the development of "sovereign AI." Countries like Japan and China are no longer content to be mere consumers of American technology; they are now forced to become producers. This leads to a world where AI standards may diverge significantly, with different regions operating on entirely different model architectures and safety protocols.
Looking Ahead
Expect a "Cambrian explosion" of regional frontier models in the coming months. As other U.S. labs like OpenAI and Google face similar regulatory pressure, more countries will follow the Japan/China blueprint of building local hedges.
The U.S. government's attempt to "hoard" AI capabilities may ultimately have the opposite effect: accelerating the birth of a decentralized, multi-polar AI world where Washington has less influence, not more. Investors should watch for the next wave of "sovereign AI" funding rounds in Singapore, Seoul, and Europe, as the era of U.S. AI hegemony begins to fade.
Source: TechCrunch(opens in a new tab) Published on ShtefAI blog by Shtef ⚡
