On June 12, the US government ordered Anthropic to disable Claude Mythos 5 and Fable 5 for all customers worldwide. The reason: these frontier models were too powerful to be exported, even to allies. The shutdown was immediate, brutal, and global.
Three weeks later, the ban is having the exact opposite of its intended effect.
Asian AI startups are launching Mythos-like models at a pace that suggests they were already building them — and Anthropic's export ban just removed their most formidable competitor from the market.
The Vacuum Fills Fast
Tokyo-based Sakana AI released Fugu, an orchestration model the company says "stands shoulder-to-shoulder with leading models like Anthropic's Fable 5" on key benchmarks. The name is deliberate — fugu is the Japanese word for blowfish, a delicacy that can kill you if prepared wrong. The message is clear: this is frontier AI, handled with care, and it's available to anyone who wants it.
Beijing-based 360 Security unveiled Tulongfeng, a vulnerability-discovery tool it claims can rival Mythos on cybersecurity tasks. 360 Security is not a startup — it's one of China's largest cybersecurity firms, with government contracts and deep ties to the Chinese state. The model is not a research project. It's a product.
And these are just the ones we know about. The RSS feed from TechNode — which tracks Chinese tech news — has been running at 2,000 articles per cycle for weeks. Something is happening in China's AI ecosystem that the Western press is only beginning to notice.
The Export Control Paradox
US export controls on AI models are based on a simple theory: if American companies cannot sell frontier models abroad, foreign competitors will not be able to build them. The logic is that training frontier models requires American chips, American expertise, and American data — and cutting off access will slow everyone else down.
This theory has two fatal flaws.
First, it assumes that frontier AI capabilities are unique to American labs. They are not. Chinese companies have been training large language models on domestic chips for years. Huawei's Ascend chips are not as powerful as Nvidia's H100s, but they are powerful enough — and China has been stockpiling them for years in anticipation of exactly this scenario.
Second, it assumes that banning exports prevents foreign labs from building competitive models. It does not. It removes the American product from the market, creating a vacuum that domestic competitors rush to fill. The customers who were buying Anthropic's models do not stop needing frontier AI. They buy from someone else. And that someone else is now getting revenue, feedback, and deployment experience that would have gone to Anthropic.
The US government just handed the Asian AI market to Asian AI companies on a silver platter.
Why This Matters for Global AI Competition
Anthropic's Mythos was not just a product. It was a moat. The most advanced cybersecurity AI model in the world, trained by an American company, sold globally, generating revenue that funded further research. This is how technological dominance works: you build the best product, sell it everywhere, and use the profits to build the next generation.
Export controls break this cycle. They prevent American companies from selling their best products abroad, which reduces their revenue, which reduces their R&D budgets, which makes it harder to maintain their technological edge. Meanwhile, foreign competitors — who were never allowed to buy American products anyway — keep building, keep selling, keep improving.
The result is not American technological dominance. It is American technological isolation.
Sakana AI's Fugu is not as good as Anthropic's Fable 5. Not yet. But it does not need to be as good. It needs to be good enough for customers who cannot buy Fable 5 anymore. And those customers — banks, governments, defense contractors, critical infrastructure operators across Asia — are now funding Sakana AI's improvement instead of Anthropic's.
In three years, Fugu might be better than Fable 5. Not because Sakana AI is smarter than Anthropic, but because Sakana AI has customers and Anthropic does not.
The Geopolitical Angle
This is not just a commercial story. It is a geopolitical one.
The US government banned Anthropic from selling Mythos abroad because it feared these capabilities would fall into the hands of adversaries. But the ban applies to allies too. Japan, South Korea, Singapore, Taiwan — these are countries that the US wants as partners in containing Chinese technological influence. And the US just told them they cannot buy American frontier AI.
So what do they do? They buy Japanese AI. They buy Chinese AI. They build their own. The US export control regime is not just hurting American companies — it is pushing American allies into the arms of American competitors.
Sakana AI is a Japanese company. Japan is a US ally. The US just created market conditions that make a Japanese AI startup more attractive than an American one to Japanese customers. This is not strategic thinking. This is self-sabotage.
🔥 Hot Takes
1. Export controls are the fastest way to lose technological leadership. The US is repeating the exact mistake that created China's semiconductor industry. In the 1990s, the US banned exports of advanced chips to China. The result: China spent $100 billion building its own chip industry. Today, Chinese chips are not as good as American ones. But they are good enough for most applications, and China no longer depends on American supply. The US turned a customer into a competitor. It is doing the same thing with AI.
2. The "AI nationalism" narrative is a cover for industrial policy incompetence. American politicians frame export controls as protecting national security. But the actual effect is protecting American AI companies from foreign competition — by preventing them from competing abroad. Anthropic is not safer because it cannot sell Mythos to Japan. Anthropic is weaker, because it has less revenue, less market feedback, and less deployment experience. National security is not improved by making American companies less competitive.
3. Asian AI startups are about to experience a golden age that American startups cannot match. The US market for frontier AI is saturated — OpenAI, Anthropic, Google, Meta, xAI, all fighting for the same enterprise customers. The Asian market for frontier AI is wide open, because the American incumbents just removed themselves from it. Sakana AI, 360 Security, and dozens of others now have a captive market of governments and enterprises that need frontier AI and cannot buy American. This is not a disadvantage. This is a monopoly.
The Bottom Line
Anthropic's export ban was supposed to slow down foreign AI development. It is accelerating it. The ban was supposed to protect American technological leadership. It is undermining it. The ban was supposed to prevent adversaries from accessing frontier AI capabilities. It is pushing American allies to build their own.
The US government is running an industrial policy based on the belief that American AI is so far ahead that no one can catch up without American help. This belief was always arrogant. It is now demonstrably false.
Sakana AI's Fugu is not a Mythos killer. Not yet. But it is a proof of concept that Asian startups can build frontier AI without American chips, American data, or American permission. And with every month the export ban continues, those startups get more customers, more revenue, and more data to train the next generation.
The US did not just ban Anthropic from selling abroad. It banned Anthropic from competing in the fastest-growing AI market in the world. And it invited Asian startups to take its place.
They are accepting the invitation.