Anthropic bans companies majority-controlled by China, Russia, Iran, North Korea (the-decoder.com)

Anthropic has tightened its terms of service to immediately bar companies that are majority‑controlled (directly or indirectly >50% ownership) by entities based in China, Russia, Iran or North Korea from using its Claude models. The change expands existing geographic and export‑control restrictions and explicitly closes a loophole that allowed firms in authoritarian states to access US models via overseas subsidiaries (examples cited include Chinese subsidiaries in Singapore). The ban covers direct customers and organizations accessing Claude through cloud providers; Anthropic warns it may forgo "low hundreds of millions" in revenue and expects some business to flow to competitors. The move is significant for the AI community because it treats ownership and jurisdiction as a national‑security risk factor: Anthropic cites the legal ability of authoritarian regimes to compel data sharing or co‑opt private firms, and the danger that access to advanced models could accelerate rival military/INTEL capabilities or speed up model cloning and distillation. Technically, the policy aims to limit transfer of capabilities that would benefit state actors until domestic hardware (or access to advanced accelerators like high‑end Nvidia GPUs, which face export limits) makes large‑scale training feasible locally. Practically, the ban primarily tightens commercial controls rather than blockading Chinese AI development, which already relies increasingly on strong homegrown models.