Analysis
As Anthropic's strongest models remain locked behind a US government export ban that bars non-Americans from accessing them, a wave of AI startups across Asia is moving to launch their own Mythos-style models -- systems pitched at the same cybersecurity-grade capability tier -- to serve the customers Washington has effectively walled off, according to TechCrunch. The episode is a live case study in how supply restrictions reshape a market.
The backdrop is the new US regime for frontier AI. Over recent weeks the administration banned non-US access to Anthropic's Mythos and the more protected Fable 5 after researchers bypassed their guardrails, then cleared Mythos for a vetted list of American agencies and firms. The result is a sharp asymmetry: the most capable American security models are available to a curated domestic list and almost no one abroad -- a vacuum that international buyers, from enterprises to governments, still need filled.
“The competitive and strategic implications cut against the policy's intent.”
That is the opening Asian labs are sprinting toward. Rather than wait for US policy to loosen, regional players are building and shipping comparable models for the non-American market, betting that customers locked out of Claude-tier systems will standardize on whatever credible alternative arrives first. It echoes the dynamic that made DeepSeek a global force -- when access to the frontier tightens or gets expensive, open-weight and foreign challengers rush the gap on cost and availability.
The competitive and strategic implications cut against the policy's intent. Export controls are designed to deny adversaries advanced capability, but in software -- where talent is mobile and architectures are widely published -- they also hand foreign firms a captive customer base and a reason to accelerate. American labs including Anthropic and OpenAI, both preparing for public listings, suddenly face a structural cap on international revenue precisely where their growth narrative is most ambitious, while rivals abroad win durable customer relationships.
The bear case is that capability is not commodity: matching a frontier lab's safety, reliability and raw performance is genuinely hard, and many of the new entrants may ship weaker models that fail to truly substitute for the real thing. What to watch: which Asian models gain real traction with serious customers, whether Washington tightens or loosens the export regime in response, and how much international revenue American labs concede before any policy adjustment arrives.
