Asian AI startups are aggressively stepping into the technological vacuum created by US export controls on Anthropic’s Mythos and Fable models, launching sophisticated, sovereign systems that signal a rapid and irreversible decentralisation of frontier AI capability. This isn’t localised competition anymore — it’s a structural shift where Washington’s geopolitical restrictions are accelerating the development of powerful, nationally-controlled AI outside the Western bloc.
🔍 THE BOTTOM LINE
The US export ban was supposed to slow Asian AI down. It did the opposite. From Beijing’s 360 group to Tokyo’s Sakana AI, competitors are now shipping models explicitly framed as Mythos-class — and in cybersecurity, 360 appears to have leapfrogged rather than caught up. The era of “all-access” frontier AI is over, and Five Eyes partners like New Zealand are caught in the middle of a fragmentation they didn’t choose.
What Changed: The Vacuum Is Now Real
The landscape shifted roughly two weeks ago when US export orders took effect, creating immediate uncertainty around access to cutting-edge compute and the foundational Mythos/Fable model families covered by the trusted-partner framework. While Western giants scrambled to update their compliance teams, Asian players did what they always do under pressure: they shipped.
The market response has been swift and uncommonly targeted. We’re not seeing general-purpose chatbot clones. We’re seeing specialised, high-capability systems designed for critical national infrastructure — cybersecurity defence, vulnerability discovery, secure orchestration. As TechCrunch reports on the Mythos-class launches, the timing is “almost too on the nose.” That sort of alignment with US policy windows is not coincidence. It’s a procurement signal.
Two Approaches: Defence vs. Frontier Capability
On one side, China’s 360 group has unveiled two tools — Tulongfeng (“Assistance Wind”) and Yitianzhen (“Sky-Scanning Needle”) — explicitly framed around national security needs. Founder Zhou Hongyi called vulnerability-finding AI a “national strategic asset” and warned of “one-way transparency” if Western adversaries retain Mythos-class capabilities while Asian defenders operate with degraded tooling. Their pitch is sovereignty-first: integrate deeply into state cyber defence, harden domestic infrastructure, and stop pretending the export ban is neutral.
On the other side, Sakana AI in Tokyo is taking the commercial route. Co-founded by ex-Google researchers Ren Ito and David Ha, Sakana launched its Fugu model this week, claiming it “stands shoulder-to-shoulder with Anthropic’s Fable 5 and Mythos Preview.” Their marketing is overtly market-facing: “frontier capability without the risk of export controls.” A Sakana spokesperson insisted the timing was “entirely coincidental.” It wasn’t. The narrative is too clean to be accident.
The Irony: The Ban as Accelerant
Controls designed to maintain Western technological hegemony are instead acting as a launchpad. Instead of creating a stable, predictable ecosystem, the export order has created a high-stakes vacuum that Asian competitors are filling with demonstrable, deployable technology. The Hacker News discussion thread on the Sakana launch skews sceptical, but even the sceptics concede the structural shift: US policy just announced to the world that frontier AI is a national security asset, and every other nation with competent AI labs is now reading the same memo.
This echoes what Ren Ito said at the G7 summit in Evian last month — that the US should “preserve access” for allies. David Ha’s framing is sharper: “Orchestration Models are the next frontier. Access to top models can disappear overnight. Collective intelligence is the practical hedge.” Sakana isn’t selling a better chatbot. It’s selling a hedge.
NZ Angle: Caught Between Two Blocs
For New Zealand, the implications are uncomfortable. As a Five Eyes ally, we benefit from security partnerships and intelligence sharing — but we have zero seat at the table when Washington decides which models get exported. The Glasswing access arrangement helped on paper, but the underlying dynamic hasn’t changed: the most advanced AI capability is now subject to sudden geopolitical whims.
For Kiwi startups, researchers, and government agencies relying on US or UK frontier access, the strategic answer is diversification — not waiting for Washington to be generous. Sovereign compute, open-weight alternatives, and direct Asian partnerships aren’t ideological choices anymore. They’re insurance.
The Other Side: The Anthropic Dilemma
Anthropic’s run-rate revenue hit roughly $47B in May 2026, and the market expects continuous, uninterrupted scaling. The export controls introduce an existential question the company has not answered cleanly: how do you maintain “frontier” status when your best customers in 80 countries can no longer buy your best models? The Reuters technology coverage of the Mythos rollout makes clear that Washington has the leverage, not Anthropic — and Anthropic’s silence on a New Zealand access path is deafening.
❓ FAQ
Q1: What are Mythos and Fable, exactly? A: They’re Anthropic’s most advanced cybersecurity-focused AI model families — Mythos is the frontier-tier system, Fable is its companion. Both are now restricted under US export controls and only available to a vetted list of “trusted partners,” which doesn’t include New Zealand startups by default.
Q2: Is this a permanent split in AI development? A: The evidence points toward regionalisation as the new baseline. We’re moving from a single global standard (mostly OpenAI/Anthropic APIs) to multiple competing ecosystems with limited interoperability. “Permanent” is too strong, but the next 2–3 years are going to look more like telecoms in the 1990s than the open web of the 2010s.
Q3: What does “Orchestration Models” actually mean? A: Systems that coordinate multiple specialised AI agents — each potentially from a different vendor or country — to solve complex, multi-step problems. It’s the architecture Sakana and others are pushing as a hedge against single-provider dependency, and it’s why “just use Claude” is no longer a viable long-term strategy.
Q4: How quickly can Asian startups actually catch up? A: They’re not catching up — they’re launching in parallel. In narrow domains like cyber vulnerability discovery, 360’s Tulongfeng may already match or exceed Mythos for specific tasks. In general reasoning and coding, the gap is narrower than most Western analysts assumed six months ago.
Q5: What should New Zealand companies do right now? A: Three things. First, audit your dependency on US frontier APIs and model the cost of sudden access loss. Second, invest in open-weight models (Llama 4, Qwen 3, DeepSeek V4) for workloads where sovereignty matters. Third, establish direct relationships with at least one Asian AI lab — Sakana, 360, or the Korean entrants — before you need them.
🔍 THE BOTTOM LINE
The narrative has decisively shifted from “who can build the biggest model” to “who can guarantee reliable access to a top-tier model.” US export controls have inadvertently handed Asian rivals the perfect marketing story: that world-class AI capability is now a distributed commodity, not a Western monopoly. New Zealand’s job in the next 24 months is to make sure we’re customers of the distributed version — not bystanders to a fragmentation we helped neither cause nor choose.
