Sakana and 360 turn Anthropic's Mythos ban into an opening

Sakana AI released Fugu, a model-orchestration product claiming frontier-level performance without single-vendor exposure, and 360 unveiled Tulongfeng as China's answer to Mythos, both capitalizing on Anthropic's suspension of Fable 5 and Mythos 5 access for foreign nationals due to a U.S. government directive. The moves highlight a market opening for Asian AI buyers seeking systems immune to U.S. export controls.

David Ha @hardmaru https://x.com/hardmaru?ref=runtimewire , Ren Ito and Llion Jones built Sakana AI around a bet that Japan did not need to copy the U.S. frontier model race to matter in AI. In the week of June 22, that bet became newly commercial: Sakana released Fugu https://sakana.ai/fugu-release/?ref=runtimewire , a model-orchestration product it says can deliver frontier-level performance without the single-vendor exposure now hanging over Anthropic's restricted Fable 5 and Mythos 5 systems. The launch landed in the same week that 360 founder Zhou Hongyi unveiled Tulongfeng, an AI vulnerability-discovery tool he positioned as China's answer to Mythos, according to TechCrunch https://techcrunch.com/2026/06/27/asian-ai-startups-launch-mythos-like-models-as-anthropics-export-ban-drags-on/?ref=runtimewire and the underlying Reuters report https://www.reuters.com/legal/litigation/chinas-360-says-it-has-developed-tools-match-anthropics-mythos-2026-06-24/?ref=runtimewire . The timing matters because Anthropic said on June 12 that the U.S. government had directed it to suspend access to Fable 5 and Mythos 5 for any foreign national, including Anthropic's own foreign national employees, forcing the company to disable the models for all customers while it complied with the order Anthropic statement https://www.anthropic.com/news/fable-mythos-access?ref=runtimewire . That is the market opening. Anthropic framed the order as a disputed safety intervention around a narrow jailbreak claim. Sakana and 360 are treating the same event as evidence that buyers in Asia need capable local or multi-provider systems that cannot be cut off by a U.S. directive. Sakana's founder bet is orchestration, not isolation Sakana's Fugu is not simply another national model pitched as a one-for-one substitute for Claude. The Tokyo startup says Fugu is itself a language model trained to call and coordinate a pool of other models. In the company's framing, the product hides the complexity of model selection, delegation, verification and synthesis from the developer using it. That architecture tracks the founding story Ha, Ito and Jones have been selling since 2023. The three former Google researchers started Sakana to build smaller, cheaper systems optimized for Japanese language and culture rather than chase the same monolithic scale curve as OpenAI, Google and Anthropic. When Sakana raised a $135 million Series B in November 2025 at a $2.65 billion post-money valuation, TechCrunch reported the company would continue building models for Japan. Fugu turns that thesis into a geopolitical product. Sakana's release page says Fugu Ultra stands "shoulder-to-shoulder" with Anthropic's Fable 5 and Mythos Preview across engineering, scientific and reasoning benchmarks while offering "frontier capability without the risk of export controls." The independent evidence is thinner than the marketing. Sakana has published a technical report on arXiv https://arxiv.org/abs/2606.21228?ref=runtimewire , which describes an agentic setup intended to adapt to queries and coordinate tools and models. But the headline comparison with Mythos and Fable remains Sakana's benchmark presentation, and Sakana notes that baseline scores for other models are reported by those model providers. 360 is making the harder sovereignty claim Zhou's message is less hedged. On June 24, 360 Security Technology introduced Tulongfeng for vulnerability discovery and Yitianzhen for automated defense and incident response, according to SC Media and TechRadar summaries of the Reuters report https://www.reuters.com/legal/litigation/chinas-360-says-it-has-developed-tools-match-anthropics-mythos-2026-06-24/?ref=runtimewire . These are company claims, not independent performance results. The South China Morning Post has reported that Zhou has warned China needs its own Mythos-like model, underscoring a sovereignty framing that goes beyond procurement risk. That is a different strategy from Sakana's. Sakana is selling resilience through access to a swappable pool of models. 360 is selling national capability in a domain where being dependent on a foreign model is not just a procurement risk, but a security liability. Anthropic's safety case created the distribution problem Anthropic's Dario and Daniela Amodei built Anthropic as a public benefit corporation around the premise that frontier capability and safety controls have to move together. That premise is now being tested in the market, not just in Washington. The company said the June 12 order cited national security authorities but did not give specific details of the government's concern. Anthropic said it understood the government's concern to involve a potential method of bypassing Fable 5 safeguards, and argued that the demonstrated vulnerabilities were minor, previously known and discoverable by other public models company statement https://www.anthropic.com/news/fable-mythos-access?ref=runtimewire . That distinction sits at the center of Anthropic's fight with the government. RuntimeWire reported on June 13 /article/anthropic-fable-mythos-jailbreak-export-control that the company said the prompt behind the shutdown was code review, not a broad cyberattack jailbreak. RuntimeWire reported on June 16 /article/anthropic-fable-mythos-cybersecurity-backlash that security leaders warned the restrictions could punish the same bug-finding work defenders need. By June 17, RuntimeWire reported /article/anthropic-nicholas-carlini-mythos-government-ai-safety that Nicholas Carlini, who had warned Anthropic not to release Mythos in March, had become central to the company's argument that guarded access was safer than a ban. The Asian launches sharpen the cost of losing that argument. A ban meant to reduce access to sensitive capability can also accelerate demand for substitutes beyond U.S. control. Sakana's spokesperson told TechCrunch the Fugu timing was "entirely coincidental" and that the underlying research had been in progress since last year and presented at ICLR in spring 2026. That can be true while the commercial effect is still obvious: Anthropic's loss of global availability is now part of Sakana's sales story. The unanswered question is performance The danger in the current cycle is that every restricted U.S. model becomes a marketing asset for someone else's launch. "Mythos-like" is a powerful phrase because Mythos has become both a product and a symbol: an AI system that defenders want, governments fear and rivals can invoke without proving equivalence. Sakana has the cleaner technical claim because it has published a report, a product page and a concrete API model. 360 has the sharper national-security claim because vulnerability discovery is closer to the export-control dispute that hit Anthropic. Neither has yet produced the kind of independently validated comparison that would settle whether Fugu Ultra or Tulongfeng actually matches Mythos or Fable 5 in the conditions customers care about. That gap does not make the launches unimportant. It makes them more instructive. The first response to Anthropic's export fight is not a single rival model beating Mythos in a public benchmark. It is founders and companies turning access uncertainty into product strategy. Ha is selling orchestration as insurance against supplier concentration. Zhou is selling domestic cyber capability as deterrence. Anthropic is still arguing that controlled deployment is safer than a recall. For buyers in Asia, the practical question is no longer whether U.S. frontier models are better on paper. It is whether the best model is still the best choice if access can disappear at 5:21 p.m. Eastern Time on a Thursday.