Blab AI
Japan and China Launch AI Models to Replace Anthropics Restricted Systems
In June 2026, a geopolitical shift in the AI arena spurred two Asian tech firms to unveil new models that promise to step into the void left by the United States’ export ban on Anthropic’s high‑performance systems. Tokyo‑based Sakana AI rolled out Fugu, a multi‑agent orchestration engine, while Beijing‑listed cybersecurity company 360 Security Technology introduced Tulongfeng, an AI tool aimed at hunting software vulnerabilities.
The U.S. government’s restriction, announced in mid‑June, barred non‑American users from accessing Anthropic’s Mythos 5 and Fable 5 models. Anthropic, headquartered in San Francisco and reporting annual revenue exceeding $47 billion as of May 2026, complied by suspending global access to the two models. The ban was justified as a national‑security measure and has prompted local competitors to develop alternatives.
360 Security Technology’s Tulongfeng is positioned as a domestic answer to Anthropic’s Mythos. The company claims the model can automatically scan code for security gaps and has been paired with Yitianzhen, a tool that automates cyber‑defense and incident response. Founder Zhou Hongyi described AI vulnerability searchers as a national strategic asset and warned of a “one‑way transparency” risk, where only a few parties possess advanced detection capabilities.
Sakana AI, founded in 2023 by former Google researchers Ren Ito, Llion Jones, and David Ha, unveiled Fugu in late June. The model is billed as a “frontier” system that rivals Anthropic’s Fable 5 and Mythos Preview in performance. Fugu is engineered to coordinate multiple specialized AI agents through a single API, allowing it to regulate access to other models. According to the company’s website, the launch coincided with the export ban but was not a direct response to it. The startup targets Japanese businesses and government agencies that need to mitigate risks from tightening export controls.
Both Tulongfeng and Fugu emphasize local language support and cultural nuance. Sakana AI notes that its models are optimized for Japanese languages and contexts, while 360’s tools are tailored to Chinese regulatory and technical environments. TechCrunch reported that these local alternatives are beginning to fill the space previously dominated by Anthropic’s models, especially for enterprises that require compliance with regional data‑protection rules.
Anthropic’s market position remains strong, with a valuation near $965 billion in May 2026. However, the company has not disclosed the size of its Asian customer base. The export restrictions have therefore created an opportunity for regional players to capture demand for advanced AI capabilities in cybersecurity and enterprise applications.
The current landscape shows a rapid shift toward domestically developed AI systems that can operate within local regulatory frameworks. While Sakana AI and 360 Security Technology have launched competing models, it remains unclear how quickly they will gain traction against Anthropic’s established infrastructure. The U.S. export ban is still in effect, and no indication has been given that it will be lifted or modified.
In the coming months, observers will watch how these new models perform in real‑world deployments, whether they achieve the performance levels claimed by their developers, and how they navigate the complex web of export controls and cybersecurity regulations.
The U.S. government’s restriction, announced in mid‑June, barred non‑American users from accessing Anthropic’s Mythos 5 and Fable 5 models. Anthropic, headquartered in San Francisco and reporting annual revenue exceeding $47 billion as of May 2026, complied by suspending global access to the two models. The ban was justified as a national‑security measure and has prompted local competitors to develop alternatives.
360 Security Technology’s Tulongfeng is positioned as a domestic answer to Anthropic’s Mythos. The company claims the model can automatically scan code for security gaps and has been paired with Yitianzhen, a tool that automates cyber‑defense and incident response. Founder Zhou Hongyi described AI vulnerability searchers as a national strategic asset and warned of a “one‑way transparency” risk, where only a few parties possess advanced detection capabilities.
Sakana AI, founded in 2023 by former Google researchers Ren Ito, Llion Jones, and David Ha, unveiled Fugu in late June. The model is billed as a “frontier” system that rivals Anthropic’s Fable 5 and Mythos Preview in performance. Fugu is engineered to coordinate multiple specialized AI agents through a single API, allowing it to regulate access to other models. According to the company’s website, the launch coincided with the export ban but was not a direct response to it. The startup targets Japanese businesses and government agencies that need to mitigate risks from tightening export controls.
Both Tulongfeng and Fugu emphasize local language support and cultural nuance. Sakana AI notes that its models are optimized for Japanese languages and contexts, while 360’s tools are tailored to Chinese regulatory and technical environments. TechCrunch reported that these local alternatives are beginning to fill the space previously dominated by Anthropic’s models, especially for enterprises that require compliance with regional data‑protection rules.
Anthropic’s market position remains strong, with a valuation near $965 billion in May 2026. However, the company has not disclosed the size of its Asian customer base. The export restrictions have therefore created an opportunity for regional players to capture demand for advanced AI capabilities in cybersecurity and enterprise applications.
The current landscape shows a rapid shift toward domestically developed AI systems that can operate within local regulatory frameworks. While Sakana AI and 360 Security Technology have launched competing models, it remains unclear how quickly they will gain traction against Anthropic’s established infrastructure. The U.S. export ban is still in effect, and no indication has been given that it will be lifted or modified.
In the coming months, observers will watch how these new models perform in real‑world deployments, whether they achieve the performance levels claimed by their developers, and how they navigate the complex web of export controls and cybersecurity regulations.
