Sakana AI and 360 Security launch rival tools as US export ban enters third week

At a glance:

• Tokyo startup Sakana AI launched Fugu, a 7-billion-parameter orchestrator that matches Anthropic's Fable 5 model on benchmarks
• China's 360 Security unveiled Tulongfeng, an AI vulnerability-discovery tool claiming to rival Anthropic's Mythos models
• The launches come as US export restrictions on Anthropic's frontier models enter their third week with no end in sight

Asian rivals step in as US tightens AI export controls

Two Asian AI companies have moved swiftly to capitalize on the vacuum created by US export restrictions on Anthropic's most capable models, with Tokyo-based Sakana AI and Beijing's 360 Security both launching products this week that directly position themselves as alternatives to suspended Mythos and Fable 5 models.

Sakana AI's approach stands apart from traditional frontier model development. Rather than training a new large language model from scratch, the company built Fugu as a seven-billion-parameter orchestrator designed to route tasks across a pool of available models, coordinating them as a team. According to the company, this approach delivers performance comparable to systems that cost significantly more to train, offering what it describes as frontier capability without the risk of export controls.

The timing of Fugu's release appears opportunistic, though Sakana has framed it as a strategic hedge. A company spokesperson told TechCrunch the timing was "entirely coincidental," but the company's website actively advertises the product as a way to preserve capability when access to single providers disappears. Co-founder David Ha went further, arguing that the export ban exposed a critical infrastructure risk: relying on one provider for national AI capabilities.

Chinese cybersecurity push accelerates amid capability gap

Meanwhile, 360 Security took a more direct approach at the ISC AI 2026 cybersecurity conference in Beijing, where founder Zhou Hongji unveiled two tools under the Yitian Tulong banner. Tulongfeng targets automated vulnerability discovery, while Yitianzhen focuses on automating cyber defense and incident response. Zhou positioned AI-driven vulnerability finding as a national strategic asset, warning of what he termed "one-way transparency"—a scenario where some countries can examine software for weaknesses while others cannot.

Despite the aggressive launch, Zhou acknowledged that Chinese AI models still trail American counterparts, estimating a 20 to 30 percent gap in base capability. However, he argued that waiting for parity was not viable, given the strategic implications. The company claims Tulongfeng has identified 3,432 software vulnerabilities to date, with 105 confirmed by Chinese authorities—a figure that Reuters could not independently verify.

Origins and backing for Sakana's orchestration approach

Sakana AI was founded in 2023 by Llion Jones, a co-author of the original Transformer paper at Google, David Ha, a former Google Brain researcher, and Ren Ito, a former Japanese diplomat. The company's research on multi-model orchestration was presented at ICLR this spring, lending academic credibility to its approach. In November 2025, Sakana raised $135 million in Series B funding at a valuation approaching $3 billion, signaling strong institutional confidence in its strategy.

The company's orchestration model represents a departure from the resource-intensive approach of training monolithic frontier models. Instead, Fugu acts as a coordinator, selecting and assembling external models to handle different aspects of complex tasks. This federated approach could prove particularly attractive to organizations seeking to avoid the concentration risks highlighted by the Anthropic export ban.

Broader Asian AI ecosystem responds

The launches reflect a wider shift across Asia in response to tightening US AI export controls. India is reportedly considering a $5 billion sovereign AI fund, while AI access dominated discussions at the recent G7 summit in Evian, France, where Sakana's Ren Ito participated among business leaders. Following the summit, Ito published an op-ed in Project Syndicate arguing that the US should preserve access for its closest allies and that AI development should be collaborative rather than hoarded.

Ito characterized the current moment as falling short of a permanent realignment, noting that "US models remain important to Asia." However, he warned that the commercial reality is shifting faster than diplomatic solutions, with enterprise customers in Asia having already begun exploring alternatives.

Commercial stakes amid ongoing restrictions

Anthropic's run-rate revenue crossed $47 billion in May, though the extent of its dependence on Asian enterprise customers remains unclear. What is evident is that two weeks of export restrictions have already produced tangible alternatives—one in an allied capital, one in a strategic rival—both built around a promise that the ban inadvertently made for them: that dependence on American AI carries risks no amount of capability can fully offset.

While neither Sakana nor 360 is claiming to replace American frontier AI outright, their launches demonstrate that the export restrictions are catalyzing real innovation in alternative architectures and deployment strategies. Whether these efforts can scale to meet enterprise demand, or whether diplomatic resolution will ease restrictions, remains to be seen.

Looking ahead

The next critical question is whether these alternative approaches can deliver consistent performance at scale. Sakana's orchestration model offers a novel path to frontier capability without massive retraining costs, but its effectiveness depends on the availability and quality of component models. Similarly, 360's vulnerability discovery claims need independent verification before broader adoption.

For policymakers, the launches underscore tensions between national security concerns and technological progress. As governments grapple with how to regulate frontier AI, these developments suggest that regional alternatives are not just theoretical possibilities but active commercial realities taking shape in real time.