After the US Export Ban: How Asia Is Reshaping the AI Landscape

In June 2026, the US government dramatically escalated AI export controls, forcing Anthropic to lock down its most powerful models from foreign nationals. The ripple effect was immediate — and it came from Asia.

Within two weeks, three major developments reshaped the competitive landscape: a Chinese cybersecurity firm unveiled a direct rival to the restricted Mythos system, a Japanese startup released an orchestration model that matches the banned US models on benchmarks without using them, and Anthropic accused Alibaba of mounting the largest-known attack on its intellectual property.

Here is a verified, source-by-source breakdown of what happened.

The US Export Ban That Started It All

On June 12, 2026, at 5:21 p.m. ET, the US Commerce Department issued an unprecedented export control directive. It ordered Anthropic to immediately suspend access to its two most advanced models — Claude Fable 5 and Claude Mythos 5 — for all foreign nationals, citing national security authorities . The ban applied regardless of location, covering any non-US person, including Anthropic's own foreign national employees .

Anthropic complied within hours, taking both models fully offline for all customers worldwide, since it had no way to selectively block only foreign nationals . Other Claude models remained unaffected.

On June 26, 2026, the US government granted a partial exception. Commerce Secretary Howard Lutnick reauthorized Mythos 5 for approximately 100 trusted companies and federal agencies with verified safeguards .

China's Answer: Tulongfeng, 'China's Version of Mythos'

At the ISC.AI 2026 conference in Beijing on June 24, 2026, Chinese cybersecurity firm 360 Security Technology unveiled two AI tools — Tulongfeng and Yitianzhen — positioned explicitly as a domestic response to Anthropic's Mythos .

Founder Zhou Hongyi called Tulongfeng "China's version of Mythos" — a vulnerability discovery and automated defense system . The company claimed Tulongfeng had already identified 3,432 software flaws, with 105 confirmed by the Chinese government .

Zhou acknowledged a 20–30% capability gap compared to Mythos but framed the tools as a strategic necessity, emphasizing the need to build a professional attack-and-defense team rather than rely on a single "genius hacker" approach .

Yitianzhen, the second tool, targets automated cyber defense and incident response, completing what 360 calls the "Yitian Tulong" suite — a name drawn from a classic Chinese martial arts novel meaning "Heavenly Sword and Dragon Saber" .

Japan's Alternative: Sakana AI's Fugu Orchestration Models

Ten days after the US ban, Tokyo-based AI startup Sakana AI (valued at approximately $2.7 billion) released Fugu and Fugu Ultra on June 22, 2026 . Rather than competing by building a larger base model, Sakana took a fundamentally different approach.

Fugu is a small ~7 billion-parameter "coordinator" model that routes tasks across a pool of publicly available frontier models — currently GPT-5.5, Gemini 3.1 Pro, and Claude Opus 4.8 . The system dynamically assigns roles: Gemini for knowledge aggregation, GPT-5.5 for mathematical reasoning, and alternating models for coding tasks .

Sakana states that Fugu's orchestrated outputs match or exceed Anthropic's restricted Fable 5 and Mythos Preview on benchmark scores, without using either restricted model — effectively sidestepping US export controls . Fugu Ultra scored 73.7 on SWE-Bench Pro, compared to Claude Opus 4.8's 69.2 and GPT-5.5's 58.6 . The system launched on OpenRouter on June 25 .

"Because the model pool is swappable, if any provider gets export-controlled overnight, Fugu just routes around it," observers noted . The company explicitly framed the timing as no accident, positioning Fugu as an alternative for any country or entity blocked from accessing US frontier models .

Anthropic Accuses Alibaba of the Largest Distillation Attack

On June 24, 2026, Anthropic sent a letter to US senators and White House officials accusing Alibaba Group and its Qwen AI lab of running "the largest known distillation attack" against Claude .

Anthropic alleged that operators used approximately 25,000 fraudulent accounts to execute over 28.8 million interactions with Claude between April 22 and June 5, 2026 . The campaign systematically extracted Claude's capabilities — particularly software engineering and agentic reasoning — to replicate them in Alibaba's own Qwen models .

The scale reportedly exceeded the combined prior distillation activity of DeepSeek, MiniMax, and Moonshot AI . Alibaba stock hit a 16-month low when the story broke . As of publication, Alibaba had not responded to the allegations .

The New Strategic Logic

Three distinct Asian responses to a single US policy reveal a new strategic logic. China moved to build domestic substitutes for restricted US cybersecurity AI, acknowledging a capability gap as a temporary challenge rather than a barrier. Japan bypassed the restriction entirely with an architectural workaround — coordination rather than competition. And through alleged distillation, some Chinese firms may have sought to shortcut the development process entirely.

The export controls on AI models, which began with chips, have now triggered a wave of innovation and conflict that will define the next phase of global AI competition.