Asian AI fills the Mythos gap: Sakana Fugu, 360 Tulongfeng, and the export-ban vacuum

Update — July 21, 2026: Sakana's Japan-enterprise strategy now has a dedicated security model: Fugu-Cyber reports 86.9% on CyberGym and 72.1% on CTI-REALM, with manual access review and human-in-the-loop verification. Read how Fugu-Cyber turns the sovereignty argument into a cyber-defense workflow →

Update — June 29, 2026 (Day 17 — vacuum persists): Fable 5: still globally banned for general users. Mythos: Annex A US critical-infra only — Asia-Pacific enterprises outside that list still blocked. Zhipu Mythos-parity security model (June 28) strengthens the case that Asian alternatives are closing the gap while US models stay permissioned. Axios June 27 reported possible Fable lift — unconfirmed; does not restore Asian access near-term. Sakana Fugu, 360 Tulongfeng, GLM-5.2, Kimi K2.7 remain the practical substitutes. Fable status. Last updated: June 29, 2026.

Update — June 28, 2026 (Austria urges EU to host Anthropic): Austria's State Secretary for Digitalization, Alexander Pröll, sent a formal letter to EU Commission Executive Vice President Henna Virkkunen urging EU member states to explore establishing Anthropic within the European Union — directly citing US restrictions on Claude Mythos and Fable as the cause. Bloomberg and Reuters confirmed. For the Asia-versus-US framing in this article, the Austria move adds a third axis: Europe is now actively building its own institutional response to the vacuum created by the ban, rather than waiting on Washington or pivoting to Asian alternatives. Whether the EU acts quickly enough to matter for Fable-dependent European developers remains to be seen. Last updated: June 28, 2026.

TL;DR — the vacuum and who is filling it

On June 27, 2026, TechCrunch reported that Asian AI companies are launching products pitched head-to-head with Claude Mythos—the cybersecurity-tier model the Trump administration treats as sensitive enough to export-control while Fable 5 stays offline for everyone else.

The headline is not one product launch. It is the second act of a story explainx.ai has been tracking since June 12: Washington pulls US frontier cyber AI offline globally; Chinese labs ship open weights within 48 hours; Japan ships orchestration; Beijing's 360 ships vuln-finding tools—with radically different politics attached.

Where the ban stands on Day 15

Anthropic launched Fable 5 and Mythos 5 on June 9. On June 12, the Commerce Department ordered a global suspension because Anthropic could not reliably block foreign-national access at API scale—a deemed-export problem explained in our international access guide.

Fifteen days later:

• Fable 5: Still banned for general users, Claude Code, and most API customers.
• Mythos 5: Partial restore only—Commerce Secretary Howard Lutnick's June 26 letter to Anthropic COO Tom Brown waives the export license for Annex A US entities, their foreign-national employees, Anthropic's foreign-national researchers, and US government labs. Everyone else needs a license. Details and letter image: Mythos trusted partners guide.
• Asia: Not on Annex A. No UK exemption survived G7 negotiations.

Anthropic told TechCrunch's Kate Park its run-rate revenue crossed $47 billion in May 2026. How much of that was Asian enterprise is unknown—but overnight access loss is now a commercial fact, not a forecast.

The story before TechCrunch: Dario's letter and the distillation war

TechCrunch frames Asia's response as a market vacuum. explainx.ai's archive shows the vacuum was policy-telegraphed days earlier.

June 10 — Dario asks for the power that gets used on him

Two days before the ban, Dario Amodei published Policy on the AI Exponential—a sweeping agenda including FAA-style authority for governments to block frontier model releases that fail independent safety testing.

Governments should have standing power to block dangerous AI deployments based on third-party evaluations.

June 12: The US blocks Fable and Mythos. The irony is documented in our full ban timeline: Dario argued for the power; Washington exercised it before Anthropic published a technical rebuttal to the jailbreak claim.

David Sacks later said the administration asked Dario to fix the jailbreak or de-deploy; Dario refused—making the standoff about process and premise, not just patches.

June 10 — The other letter: 25,000 fake accounts

Same day as Dario's essay, Sarah Heck (Anthropic Head of Policy) wrote the Senate Banking Committee about the largest known distillation attack on Claude: ~25,000 fraudulent accounts, 28.8 million exchanges (April 22 – June 5), operators linked to Alibaba / Qwen, harvesting agentic reasoning and coding toward Mythos Preview-class capability.

Full breakdown: Anthropic vs Alibaba — 25K fake accounts.

That letter matters for the Asia story because it exposes a contradiction Washington and Anthropic now own:

• Block frontier Claude from foreign nationals via export control.
• Simultaneously fight industrial-scale capability extraction into open-weight Chinese models already shipping globally.

Asian enterprises are not choosing between "US cloud" and "nothing." They are choosing between permissioned US access and unrestricted local or open stacks.

Act I — China: speed, not hedging (June 12–13)

Within 48 hours of the ban:

Zhipu co-founder Jie Tang framed the release partly as belief that "frontier intelligence belongs to everyone"—a direct response to a global cutoff, not a US-only gate.

TechCrunch's June 27 chapter adds 360 Security—a different tone entirely.

Act II — Japan: Sakana Fugu as hedge, not replacement (June 22)

Sakana AI launched Fugu on June 22—the day Anthropic's free Fable pricing window expired with zero hours delivered since June 12.

Sakana's pitch to TechCrunch:

• Research roots in ICLR 2026 (TRINITY + Conductor papers)—"entirely coincidental" timing vs the ban.
• Website copy: "frontier capability without the risk of export controls."
• Co-founder Ren Ito at G7 Évian: US models "remain important to Asia"—not a permanent realignment.
• CEO David Ha on X: "Orchestration Models are the next frontier" — "Access to top models can disappear overnight. Collective intelligence is the practical hedge against concentration of power."

That is the architectural answer to export controls: one API, many models behind it—swap providers when one jurisdiction turns off the tap.

explainx.ai's early testing notes a gap between benchmark claims (SWE Bench Pro 73.7, LiveCodeBench 93.2) and some real workloads (Ethan Mollick reported 30-minute shader runs, quality below Fable on Harbor-style tests). Treat Fugu as sovereignty infrastructure, not a drop-in Fable clone, until you validate on your tasks.

Act III — China again: 360 and "one-way transparency" (late June)

TechCrunch reports 360 unveiled two tools Wednesday (June 25 in the reporting cycle):

Founder Zhou Hongyi (via Reuters) described vuln-finding AI as a national strategic asset and warned of "one-way transparency"—some actors seeing every flaw while others cannot.

Where Sakana hedges, 360 competes narratively with Washington's Mythos framing: if Project Glasswing-class models can find NSA-scale vulnerabilities in hours, who gets that capability becomes a geopolitical question—not just a product question.

360 did not respond to TechCrunch's comment request.

The G7 thread: allies wanted access; Annex A is US-only

At G7 Évian (June 17), Dario Amodei and Ren Ito sat in the same building while Fable/Mythos were dark.

• Trump: Anthropic talks "going fine."
• Macron: Trusted Partners scheme for allied frontier access—still negotiating.
• Ren Ito (Project Syndicate): US "first priority should be to preserve access" for closest allies; "AI should not become a technology that is hoarded."

June 26 Lutnick letter: Trusted partners exist—but Annex A is a US institution list, not a Japan–UK–EU passport. The UK exemption collapsed June 17. Asia's enterprise buyers read that as: US cyber AI is for US-approved orgs.

Parallel: GPT-5.6 Sol preview the same week—permissioned US release vs open-weight GLM/Kimi/Qwen anyone can host.

Who actually wins the vacuum?

Short term:

Structural:

Satya Nadella's line after the ban—"a frontier without an ecosystem is not stable"—maps cleanly to TechCrunch's Asia beat. When one government's export control deletes global API access, orchestration and open weights become default architecture.

Anthropic may restore Fable under ID verification (July 8) or August 1 EO framework deals. Even then, trust among Asian buyers who built June 2026 contingency stacks will not reset automatically.

The uncomfortable symmetry

Washington banned Mythos partly because foreign nationals might gain offensive cyber capability.

Anthropic's June 10 letter documents foreign operators already distilling Claude at industrial scale while Fable was still live.

Asian labs now ship Mythos-class marketing without US export paperwork.

Three truths coexist:

1. Export controls are a real lever—Lutnick's letter proves Washington will use it surgically.
2. Distillation and open weights are a bypass export controls cannot fully close.
3. Orchestration (Fugu) reduces single-vendor kill-switch risk without requiring any one lab to match Mythos weights.

TechCrunch's June 27 story is the human-readable headline for that triangle.

What to watch next

Bottom line

The US did not just pause a product—it opened a regional competition for cyber-tier frontier AI.

Sakana Fugu sells resilience through orchestration. 360 Tulongfeng sells strategic parity. GLM-5.2 and Kimi K2.7 sold open access before TechCrunch's headline landed.

Dario's June 10 essay asked governments to ground models when unsafe. Lutnick's June 26 letter grounded Mythos for Annex A only. Fable 5—what most developers actually used—remains banned.

Asia is not waiting for the next letter. Neither should your architecture.

Sources: TechCrunch (Jun 27, 2026), explainx.ai coverage linked above, Reuters reporting on 360 cited by TechCrunch.