Asian AI startups step into the breach as U.S. export controls leave Mythos out of reach

Asia’s frontier AI models arrive as U.S. restrictions keep Mythos off global markets

U.S. export controls have temporarily blocked access to Anthropic’s Mythos model outside American borders, creating an immediate gap in high-performance AI availability for international users. In response, two Asian AI startups—360 in China and Sakana AI in Japan—have launched new frontier models designed to match Mythos-level capabilities while remaining outside the scope of U.S. restrictions. These moves signal a strategic shift: global organizations that previously relied on U.S.-based models are now turning to Asian alternatives to maintain access to advanced AI systems without regulatory risk. For businesses and governments outside the U.S., this development is not just about compliance—it’s about continuity in AI-driven operations.

The timing of these launches is striking. Both companies introduced their models within days of the U.S. government’s decision to restrict Mythos and its restricted variant, Fable 5, from non-American users. While Sakana AI emphasized that its Fugu model was in development for over a year and presented at ICLR earlier in 2026, the coincidence of its release during the export ban has amplified its visibility. The company’s messaging is clear: deliver frontier capability without the regulatory exposure. Meanwhile, 360’s Tulongfeng positions itself directly against Mythos in the cybersecurity domain, suggesting that Asian developers are not only replicating performance but also targeting specific high-stakes use cases that U.S. models are now restricted from serving abroad.

How Mythos-level performance is now available outside U.S. control

Anthropic’s Mythos has been positioned as a high-performance, agent-capable model optimized for complex workflows, including cybersecurity and enterprise automation. The U.S. government’s decision to restrict its global distribution has left organizations outside the country without direct access to a model of comparable caliber—at least temporarily. Tulongfeng from 360 and Fugu from Sakana AI are explicitly designed to fill this void. Tulongfeng is described as a cybersecurity-focused AI tool capable of going “head-to-head” with Mythos, while Fugu is positioned as a frontier AI model that “stands shoulder-to-shoulder” with leading systems like Fable 5 and Mythos Preview.

What makes these new models significant is their dual focus: high performance and regulatory compliance. Both companies acknowledge that their models are built for agentic behavior—meaning they can orchestrate other AI systems via APIs, a feature central to modern enterprise automation. This capability is not merely technical; it reflects a growing demand for AI systems that can integrate with existing toolchains without introducing legal or geopolitical risk. For organizations in regulated industries—such as finance, healthcare, or government—this alignment between capability and compliance is a critical advantage.

The geopolitical dimension: export controls and the fragmentation of AI supply chains

The U.S. export ban on Mythos and Fable 5 represents a turning point in the global AI landscape. It is the first major instance of a Western AI model being restricted from international distribution due to national security concerns, setting a precedent that could reshape supply chains. Historically, U.S. AI models have dominated global markets, but the imposition of export controls introduces a new variable: uncertainty. Organizations that have built their AI strategies around U.S.-developed models now face potential disruptions, compliance audits, and the need to diversify their supplier base.

This regulatory shift is accelerating the rise of non-U.S. AI ecosystems. Japan’s Sakana AI, co-founded by former Google researchers, exemplifies this trend. The company’s focus on affordability, efficiency with small datasets, and cultural optimization for Japanese language and context positions it as a natural alternative for enterprises and public agencies in the region. Similarly, 360’s entry into the frontier AI space with Tulongfeng underscores China’s ambition to develop homegrown models that can serve domestic and international markets without U.S. restrictions. The result is a more fragmented AI supply chain, where regional players are gaining influence and market share by offering compliant, high-performance alternatives.

What these models mean for enterprise and government users

For enterprises outside the U.S., the emergence of Tulongfeng and Fugu offers a practical solution to a pressing problem: how to maintain access to advanced AI capabilities without violating export regulations. Cybersecurity teams, in particular, may find Tulongfeng compelling given its direct positioning against Mythos in security-focused tasks. Meanwhile, organizations looking for agentic AI systems that can integrate with multiple APIs—such as customer support platforms or internal automation tools—may turn to Fugu for its orchestration capabilities. The message from both companies is consistent: these models are not just alternatives; they are designed to meet the same performance benchmarks as the restricted U.S. models.

Government agencies, especially in Japan and other Asia-Pacific nations, are likely to be early adopters. These entities often operate under strict data sovereignty and security requirements, making compliance with U.S. export controls a significant barrier. By adopting models developed and hosted within their own jurisdictions, agencies can avoid the risk of unintended exposure to U.S. restrictions. This shift could also influence procurement policies, with governments prioritizing AI systems that are developed locally or regionally to ensure long-term access and control.

The technical underpinnings: efficiency, language optimization, and agentic workflows

Sakana AI’s Fugu model is built on research presented at ICLR 2026, indicating a foundation in cutting-edge AI techniques. The company’s focus on efficiency with small datasets and optimization for Japanese language and cultural context suggests a pragmatic approach to AI deployment. This is particularly relevant for organizations that need high performance but operate within limited data environments or require culturally nuanced outputs. The model’s agentic capabilities—its ability to orchestrate other AI systems—align with the growing trend of AI-driven automation in enterprise settings.

360’s Tulongfeng, on the other hand, is positioned as a cybersecurity-focused AI tool. This specialization reflects a strategic choice to target a high-value, high-need domain where U.S. models are now restricted. Cybersecurity is an area where AI can significantly enhance threat detection, incident response, and compliance monitoring. By offering a model that can compete with Mythos in this space, 360 is not only filling a technical gap but also addressing a critical operational need for organizations that can no longer rely on restricted U.S. systems.

Market implications: can Asian models capture U.S. market share?

While Tulongfeng and Fugu are currently positioned for non-U.S. markets, the long-term implications for U.S. AI companies are significant. If these models gain traction and prove capable of matching or exceeding the performance of restricted U.S. models, they could erode the dominance of American AI labs in global markets. This is especially true in regions where geopolitical tensions or regulatory concerns make U.S. models less attractive. The export ban may have inadvertently accelerated the rise of alternative AI ecosystems, giving Asian developers an opportunity to establish themselves as credible competitors.

For U.S. AI labs like Anthropic, the challenge is twofold: navigating evolving export controls while maintaining global competitiveness. The restrictions on Mythos and Fable 5 may force these companies to reconsider their international strategies, potentially leading to partnerships, localized deployments, or even the development of compliant variants. However, the window for U.S. models to regain unconstrained global access may be narrowing as Asian alternatives mature and gain trust. Organizations that once defaulted to U.S. AI systems may now evaluate regional options more seriously, driven by both performance and compliance considerations.

What organizations should watch next

The immediate priority for organizations is to assess the capabilities of Tulongfeng and Fugu in their specific use cases. Enterprises should evaluate whether these models meet their performance requirements for tasks such as cybersecurity, automation, or data analysis. Pilot programs and benchmarking against existing systems will be essential to determine feasibility. Additionally, organizations must consider the legal and operational implications of switching to non-U.S. models, including data residency, export compliance, and vendor lock-in risks.

Another critical area to monitor is the response from U.S. AI labs. Will they introduce compliant variants of Mythos or Fable 5 for international markets? Alternatively, will they accelerate partnerships with local providers to ensure continued access? The outcome will shape the future of AI supply chains and influence how organizations balance performance, cost, and compliance in their AI strategies. For now, the message is clear: the AI landscape is diversifying, and the era of unchallenged U.S. dominance in frontier AI may be giving way to a more multipolar ecosystem.

Conclusion

The launch of Tulongfeng and Fugu marks a pivotal moment in the AI industry, highlighting how export controls can reshape global markets and accelerate the rise of regional alternatives. These models offer organizations outside the U.S. a viable path to access frontier AI capabilities without violating restrictions, while also signaling a broader shift toward a more fragmented and competitive AI landscape. For enterprises and governments, the message is to act swiftly: evaluate these new options, pilot their integration, and prepare for a future where AI supply chains are no longer dominated by a single region. The age of unconstrained global AI access may be over, but the age of choice—and competition—has only just begun.