A U.S. export-control order that forced Anthropic to pull its two most powerful artificial-intelligence models offline is reshaping the competitive map far beyond Silicon Valley. In the weeks since the directive took effect, at least two Asian companies, one in Beijing, one in Tokyo have rushed out cybersecurity-capable models pitched explicitly as alternatives to Anthropic's restricted "Mythos-class" systems.
The launches are an early, concrete sign that Washington's attempt to keep its most advanced AI out of foreign hands may be accelerating the very thing it set out to prevent: homegrown rivals with comparable capabilities and no exposure to American export rules.
The ban that started it
On June 12, the U.S. government, citing national-security authorities, issued an export-control directive ordering Anthropic to suspend all access to its Fable 5 and Mythos 5 models for any foreign national inside or outside the United States, and including the company's own non-citizen employees. Because the order swept so broadly, Anthropic said it had little choice but to switch the models off for everyone to stay compliant. Access to its other Claude models, including the latest Claude Opus 4.8, was not affected.
Mythos, previewed in April, is a system built to autonomously hunt for software flaws; Anthropic said an early version had uncovered "thousands" of serious vulnerabilities across operating systems, browsers and other applications. Fable 5 is a tuned, public-facing version of that technology. The government pointed to a method of "jailbreaking" Fable 5's safeguards, but Anthropic pushed back hard, describing it as a narrow, non-universal flaw that other publicly available models including OpenAI's GPT-5.5 also exhibit. "We believe this is a misunderstanding," the company said in a statement, adding that it was working to restore access as soon as possible.
The shutdown did not happen in a vacuum. It capped months of friction between Anthropic and the Trump administration, which earlier this year barred federal agencies from using the company's models after a contract dispute and later labeled Anthropic a "supply chain risk" , a designation the company is challenging in federal court. Cybersecurity researchers also criticized the order, arguing that vulnerability discovery is now common across many foundation and open-source models and that singling out one vendor made little technical sense.
China's answer: 360's 'Tulongfeng'
The most direct response came in Beijing. At the ISC.AI 2026 cybersecurity conference, 360 Security Technology founder Zhou Hongyi unveiled two AI security tools under the banner "Yitian Tulong," a reference to a classic martial-arts novel translated as "Heavenly Sword and Dragon Saber." One, Tulongfeng, automates software-vulnerability discovery; Zhou called it "China's version of Mythos." The other, Yitianzhen, is built to automate cyber defense and incident response.
Zhou was unusually candid that China's underlying models still trail their American counterparts. "Objectively speaking, domestic models still have a 20%-30% gap in base capability," he said, according to Reuters. His workaround is engineering rather than raw scale layering AI agents on top of 360's vulnerability databases and two decades of security expertise. "If the U.S. route is to cultivate a genius hacker, 360's route is to organise a professional attack-and-defence team," he said.
He cast the project in stark strategic terms, warning of "one-way transparency" a world in which American tools could probe foreign systems at will while others could not. "This kind of powerful weapon … cannot be held only by others," Zhou said, according to a transcript published by 360. The firm claimed Tulongfeng had already identified 3,432 software vulnerabilities, 105 of them confirmed by Chinese authorities. Zhou, a veteran internet entrepreneur and a member of China's top political advisory body, has built 360 into one of the country's best-known security brands.
Tokyo's hedge: Sakana's 'Fugu'
Days earlier, Tokyo-based Sakana AI founded in 2023 by former Google researchers Llion Jones, David Ha and Ren Ito released Fugu, named after the Japanese blowfish. The company says the frontier model "stands shoulder-to-shoulder" with Anthropic's Fable 5 and Mythos Preview, and its website pitches "frontier capability without the risk of export controls."
Unlike 360, Sakana frames Fugu less as a replacement than as insurance. The model is built for orchestration coordinating other models through their APIs rather than trying to outscale them. "Orchestration Models are the next frontier, beyond bigger models," chief executive David Ha wrote on X. Leaning on a single provider for national infrastructure, he argued, had become a risk impossible to ignore: "Access to top models can disappear overnight."
Even so, Sakana has stopped short of declaring the end of American AI in Asia. Co-founder Ren Ito, writing in Project Syndicate, urged Washington to "preserve access" for its closest allies and argued that "AI should not become a technology that is hoarded." The company has also characterized the timing of Fugu's release as coincidental, even as it has leaned into the attention the moment has brought. The model is aimed at Japanese businesses and government agencies looking to reduce their exposure to tightening export controls.
A partial thaw but the gap is already filling
The pressure now appears to be easing. On June 27, Anthropic said the government had cleared it to redeploy Mythos 5 to a set of U.S. organizations that "operate and defend critical infrastructure," with Reuters reporting that more than 100 companies and institutions would regain access. Commerce Secretary Howard Lutnick, in a letter shared with the press, said the two sides' work had "yielded significant progress." By the weekend, people at Anthropic told Axios that Fable 5 was expected to return to worldwide availability soon potentially within the week though the company has offered no firm timeline.
Yet even a full reversal may not undo the damage. Anthropic, which said its run-rate revenue crossed $47 billion in May and has confidentially filed to go public, has never disclosed how much of that business depends on Asian enterprise customers. What is clear is that the vacuum did not stay empty. Models tuned for local languages, local rules and local control are now on the market and trust, once shaken by an overnight shutoff, is far harder to win back than access.
The episode has supercharged a debate over "AI sovereignty" a nation's ability to control its own artificial intelligence. British and other policymakers have already cited the ban as a reason to invest more heavily at home. For governments that just watched the world's most advanced models go dark on a few hours' notice, the lesson of June 2026 is blunt: dependence is a liability, and the alternatives are arriving faster than almost anyone expected.
Comments