72 hours.

On June 9, Anthropic surprise-launched two top-tier models -- Claude Fable 5 for the public and Claude Mythos 5 for select B2B partners. They share underlying weights, differing only in safety guardrails, and broke records across multiple benchmarks.

72 hours later on June 12, the White House issued an export control order on Fable 5 and Mythos 5 on national security grounds, citing concerns that the models could be jailbroken for malicious purposes.

This is the first time in human history that an AI model has been placed under strategic-materials-level export controls less than three days after launch. The historical significance of this event can hardly be overstated.

Why 72 Hours?

The export controls came so fast that they indicate the U.S. government had already prepared assessments of Anthropic's new model capabilities. Fable 5 is the first Mythos-level model released to the public -- "Mythos" is Anthropic's internal designation for its highest capability tier, never before opened to the public.

The core justification for controls is "jailbreak risk": the U.S. government fears that if Mythos-level models were open-sourced or obtained by foreign entities, they could be used for cyberattacks, bioweapons design, large-scale disinformation generation, and other high-risk scenarios. But the deeper reason is the generational capability gap in frontier AI -- when model capability crosses a certain threshold, it ceases to be merely a commercial product and becomes a core component of national competitiveness.

Canadian Prime Minister Mark Carney subsequently publicly warned about the risks of over-reliance on a handful of American AI providers. Indian commentary also framed this as a warning signal about long-term dependence on U.S. AI infrastructure. The reactions from these two countries show that U.S. export controls haven't reassured allies -- they've accelerated global thinking about "de-Americanization" of AI supply chains.

The Unexpected Alliance of Silicon Valley's Big Three

Another signal deserves more attention: Anthropic sued Alibaba over model distillation, and revealed it has shared a database of violating distillation cases with OpenAI and Google. This is the first time these three companies -- fierce competitors on the U.S. AI agenda -- have formed a substantive alliance on IP protection and export controls.

The emergence of this alliance sends a clear signal: America's AI big three are aligned on "external defense." However they compete for market share commercially, they stand on the same side when it comes to preventing competitors like China from narrowing the model capability gap through distillation, reverse engineering, or similar means.

Elon Musk retorted on X that "Anthropic itself is guilty of massive training data theft," but such voices have been marginalized in mainstream Silicon Valley narratives.

The Balkanization of Global AI Markets

The long-term impact of the Anthropic export control incident will be accelerating fragmentation of the global AI market.

Enterprises and governments will have to confront a reality: depending on U.S. frontier AI models carries policy risk -- you can use them today, but a single control order tomorrow could cut you off. This will drive three trends: the EU accelerating development of domestic AI capabilities, populous nations like India increasing investment in AI sovereignty, and China continuing its independent technology path.

For Chinese AI companies, this is both pressure and motivation. Pressure: the window for accessing the most frontier model capabilities is further tightened. Motivation: when global customers recognize single-vendor risk, alternatives that don't depend on the U.S. tech stack gain greater market space. DeepSeek completing full-parameter training on domestic chips looks especially valuable against this backdrop.

The "Balkanization" of frontier AI is irreversible. The future global AI landscape won't be a flat, interconnected world, but one where several large technology ecosystems coexist and compete.