The irony is almost too neat: Anthropic spent years positioning itself as the safety-conscious alternative to OpenAI, the lab that would prove you could build powerful AI responsibly. Now that very caution—manifested in a months-long export compliance review that has effectively frozen Claude's expansion into key Asian markets—is handing market share to rivals who face no such constraints.

This week, at least three well-funded startups in Japan, South Korea, and Singapore announced or quietly deployed large language models explicitly designed to fill the Claude-shaped hole in enterprise offerings. The messaging is consistent: why wait for Washington to decide whether your business can access American AI?

The compliance trap

Anthropic's export difficulties stem from a familiar tension. The company's own safety research has documented how frontier models could, in theory, assist with weapons development or cyberattacks—findings that regulators have used to justify tighter scrutiny of international sales. The result is a bureaucratic purgatory: Claude remains technically available in most of Asia, but enterprise contracts requiring certain compliance certifications have stalled for months.

For large Asian corporations—banks, manufacturers, logistics firms—this uncertainty is untenable. They need to plan AI infrastructure years in advance. A model that might be restricted tomorrow is worse than a slightly less capable model that will definitely be available.

The Mythos playbook

The new Asian entrants are explicitly modeling themselves on Mythos, the open-weights project that demonstrated you could build near-frontier capabilities without American cloud dependencies. The approach is pragmatic rather than ideological: train on multilingual data that emphasizes local languages, partner with regional cloud providers, and price aggressively against incumbents who must factor in compliance overhead.

Early benchmarks suggest these models trail Claude 3.5 Sonnet by meaningful margins on complex reasoning tasks. But for the bread-and-butter enterprise use cases—document processing, customer service automation, code assistance—the gap is narrowing faster than most Western observers expected.

Our take

Anthropic's predicament illustrates a structural problem with American AI leadership: the same regulatory seriousness that makes US labs credible on safety also makes them slower, more expensive, and less predictable as business partners. Asian competitors do not need to match frontier capabilities—they just need to be good enough, available now, and free of geopolitical entanglements. The longer Anthropic's export review drags on, the more "good enough" starts to look like the winning strategy.