The ban that launched a thousand rivals

When the Trump administration banned Anthropic‘s Mythos and its restricted variant Fable 5 from non-American access, it framed the decision as a national security necessity — the models’ advanced cybersecurity capabilities were simply too dangerous to export. What Washington did not anticipate was that the ban would function as a public capability benchmark, broadcasting to every competing AI lab on the planet exactly where the threshold of strategic concern now sits.

The response from Asia was immediate and pointed. Within days of the export restriction making headlines, Chinese cybersecurity firm 360 unveiled Tulongfeng, explicitly positioning it as a direct competitor to Mythos. Days earlier, Tokyo-based Sakana AI launched Fugu — named after the Japanese word for blowfish — and described the frontier model as standing “shoulder-to-shoulder” with both Fable 5 and Mythos Preview. Fugu is also built for agentic workflows, capable of orchestrating other models through their APIs, targeting the same autonomous-systems use cases that made Mythos a national security concern in the first place.

Two Mythos-equivalent releases in a single week, from two different Asian markets, is not coincidence. It is a signal. The U.S. export control regime was designed to slow the diffusion of advanced AI capabilities beyond American borders. Instead, the ban handed rivals a precise capability target and the marketing framing to claim they had hit it.

The geopolitical calculus here matters. AI export controls, like semiconductor restrictions before them, assume that withholding access creates durable advantage. But model capabilities are not chips — they cannot be embargoed at a port. When Washington publicly defines what a dangerous AI model looks like, it effectively publishes a product specification. Competitors in Beijing and Tokyo read the same headlines, and Tulongfeng and Fugu are the result. The global AI power balance is shifting not despite the ban, but partly because of it.

Meet the challengers: Tulongfeng and Fugu

Chinese cybersecurity firm 360 unveiled Tulongfeng on Wednesday, framing it as a direct competitor to Anthropic’s Mythos — the AI model currently locked behind a U.S. export ban. The positioning is deliberate and the source is significant. 360 operates deep inside China’s state cybersecurity infrastructure, which means Tulongfeng is not a scrappy startup bet on frontier AI. It is a state-adjacent organization openly declaring capability parity with the most restricted American AI model on the market.

The same week, Tokyo-based Sakana AI launched Fugu. The name comes from the Japanese word for blowfish — a creature prized as a delicacy and capable of killing the person who eats it if prepared incorrectly. That dual identity is not accidental branding. Sakana claims Fugu stands shoulder-to-shoulder with both Fable 5 and Mythos Preview, and the model is built for agentic tasks, capable of orchestrating access to other AI models through their APIs. That kind of multi-model coordination puts it squarely in the territory of advanced AI deployment, not just benchmark competition.

Two frontier AI releases from two different Asian countries in the same week, both explicitly measured against Anthropic’s export-banned models, is a data point Washington cannot ignore. This is not one country filling a vacuum the U.S. created. This is a regional shift in AI development ambition playing out in real time. China is moving through established players with state ties. Japan is moving through well-funded startups with global reach. The parallel timing signals coordination of confidence, if not of strategy.

The U.S. export control framework was built on the assumption that restricting access to advanced AI models buys time and preserves technological advantage. Tulongfeng and Fugu together challenge that assumption directly. The ban on Mythos was meant to keep the most capable AI out of foreign hands. Instead, it handed foreign developers a competitive benchmark and a marketing target — and they launched against it within the same news cycle.

What most coverage is missing: the cybersecurity dimension

Most coverage frames the Anthropic export ban as a trade dispute or a chapter in the broader US-China tech rivalry. That framing misses the point. Mythos is not a general-purpose large language model — it is a cybersecurity-focused AI system, and that single fact changes the entire stakes of the story.

When the Trump administration blocked international access to Mythos and its companion model Fable 5, it was specifically trying to prevent adversaries from acquiring AI-powered cyber capabilities — tools built to identify software vulnerabilities, automate threat analysis, and potentially assist in the construction of offensive exploits. The export restriction was never really about market competition. It was about keeping a weapons-adjacent technology out of hostile hands.

Chinese cybersecurity firm 360’s decision to build Tulongfeng as a direct Mythos competitor is not a coincidence of timing. 360 operates inside China’s cybersecurity ecosystem, which has documented ties to state-sponsored threat actors. A domestically developed AI model that matches Mythos on vulnerability research and cyber reasoning benchmarks hands those actors something the export ban was explicitly designed to deny them: AI-assisted offensive and defensive cyber tooling with no dependency on American infrastructure or licensing agreements.

The implications run directly into critical infrastructure. AI-accelerated vulnerability discovery compresses the timeline between a zero-day’s existence and its weaponization. A model capable of autonomous threat detection on one side can, in adversarial hands, become an automated attack-surface scanner on the other. Neither mainstream reporting on the ban nor the diplomatic noise around it has seriously engaged with this dual-use reality.

Sakana AI’s Fugu, the Tokyo-built model positioned alongside Fable 5 and Mythos Preview, adds another layer. Its agent architecture — designed to orchestrate access to other AI models through external APIs — raises distinct questions about how interconnected AI systems could be leveraged in coordinated cyber operations. The US moved to restrict one model. The response produced at least two replacements, both built outside American jurisdiction, both arriving faster than Washington’s export control apparatus anticipated.

The export control paradox: does banning advanced AI accelerate its spread?

Washington’s decision to name Mythos explicitly in its export ban handed foreign competitors something more valuable than the model itself: a certified benchmark to chase. The moment the Trump administration publicly identified Mythos as too powerful for non-American hands, it transformed Anthropic’s model into a prestige target with a government-stamped credential attached.

The response from Asia came fast. Chinese cybersecurity firm 360 unveiled Tulongfeng within weeks, marketing it directly as a system capable of going head-to-head with Mythos. Tokyo-based Sakana AI launched Fugu — named after the Japanese blowfish — and positioned it as standing “shoulder-to-shoulder” with both Fable 5 and Mythos Preview. Neither company needed internal intelligence about Mythos’s architecture. The US government’s own restriction told them exactly which capability tier mattered most.

This pattern has precedent. Semiconductor export controls in the 1980s accelerated Japan’s domestic chip development rather than containing it. Cryptography export restrictions in the 1990s, which once classified strong encryption as a munition, pushed European and Asian developers to build independent encryption standards that eventually circulated freely regardless of American policy. In both cases, the restriction functioned as an advertisement for the technology’s strategic importance.

The dilemma facing AI policymakers now mirrors those earlier failures, with one additional complication. Keeping Mythos and Fable 5 domestic limits Anthropic’s commercial reach in markets where AI adoption is accelerating — Southeast Asia, the Gulf states, India — while simultaneously broadcasting to every competitor laboratory exactly which model capabilities the US intelligence community considers decisive. Rivals do not need to steal the model. They need only read the ban order to understand what to build next.

Export controls on AI models operate differently than controls on physical hardware. A chip fab requires rare equipment and specialized supply chains. A frontier AI model requires compute, data, and engineering talent — resources that are increasingly distributed across the countries Washington is trying to exclude from access.

What this means for Anthropic — and US AI strategy

Anthropic built Mythos to be its most capable model — and that capability is now working against it. The export ban locks the company out of international markets at exactly the moment Asian competitors are rushing to fill the vacuum. Chinese cybersecurity firm 360’s Tulongfeng and Tokyo-based Sakana AI’s Fugu aren’t waiting for Washington to sort out its policy. They are already positioning themselves as direct alternatives to Mythos and Fable 5, targeting the same enterprise and government buyers across Asia that Anthropic can no longer reach.

The commercial stakes are not abstract. Finance, healthcare, and government sectors across Southeast Asia, Japan, South Korea, and beyond represent enormous AI adoption opportunities. Enterprise AI procurement decisions tend to be sticky — once an organization builds workflows, compliance frameworks, and vendor relationships around a model, switching is expensive and disruptive. Every month the ban continues, Tulongfeng and Fugu have the opportunity to embed themselves deeper into institutional infrastructure. By the time any restriction lifts, Anthropic may be fighting for second place in markets it could have led.

The deeper problem is strategic incoherence in Washington. The US government has no publicly articulated framework that explains which AI capabilities trigger export controls, under what conditions restrictions get lifted, or how allied nations are supposed to access American frontier AI. The Mythos ban looks less like a calculated geopolitical move and more like a reactive measure — one that imposes real costs on a domestic AI company while doing little to prevent rival nations from accelerating their own frontier model development. Tulongfeng’s emergence from a Chinese cybersecurity firm is a direct illustration of that dynamic.

American AI export policy cannot simultaneously claim to promote US technological leadership and deny US companies the ability to compete internationally. The restriction on Mythos and Fable 5 forces Anthropic to absorb competitive damage while foreign AI developers face no equivalent constraint in their target markets. That asymmetry is not a side effect of the policy — it is the policy’s defining consequence.

The bigger picture: Asia’s AI moment

The launches of Tulongfeng and Fugu are not coincidences or outliers — they are signals of a maturing AI ecosystem across Asia that now has the talent, capital, and compute infrastructure to build at the frontier. Chinese firm 360 and Tokyo-based Sakana AI did not spend years waiting for Washington to define the rules. They built.

Sakana AI represents something genuinely new in the global AI landscape. Founded in Tokyo with an internationally recruited research team, it operates entirely outside the Silicon Valley orbit while pursuing frontier model development. Fugu is not a derivative product or a fine-tuned Western model — Sakana positions it as a peer to Anthropic’s own Fable 5 and Mythos Preview, with native agentic capabilities that let it orchestrate other models through their APIs. That kind of technical ambition, coming from a Japanese startup, would have seemed implausible to most American observers just three years ago.

The deeper story here is structural. For roughly a decade, advanced AI was effectively a US-origin product — something the rest of the world consumed, licensed, or tried to replicate. That era is ending. Asian AI development is no longer catch-up competition. It is parallel frontier research, funded independently, staffed with world-class researchers, and increasingly unbothered by what American export controls permit or prohibit.

Washington’s export ban on Mythos was designed to preserve US advantage in high-capability AI. What the simultaneous appearance of Tulongfeng and Fugu demonstrates is that the advantage was already narrowing before the ink dried on the order. Restricting access to a model only matters if rivals cannot build a comparable one. Right now, they can — and they are doing it faster than US policymakers anticipated. The global AI power balance is not shifting gradually. It is shifting now.