Anthropic blocks Chinese-owned firms from using Claude models under new terms

Anthropic has updated its Terms of Service to block access for companies majority-owned or controlled by Chinese entities, defined as at least 50 percent Chinese ownership, regardless of operating location.

Anthropic cites legal, regulatory, and security risks for these entities, including concerns over military or intelligence use and model distillation. The new Terms of Service took effect on the 5th of September, focusing on ownership-based restrictions rather than geography.

The restriction covers all Claude models, including Claude 3.5 Sonnet, and extends to developer tools, subsidiaries, and joint ventures. Separately, Anthropic advocates for export controls and national-security evaluations for advanced AI models.

The new policy is expected to affect ByteDance, Tencent, Alibaba, and their subsidiaries and portfolio companies, with an estimated revenue impact in the low hundreds of millions of dollars. While Chinese firms have previously faced Western technology blocks, this is the first restriction based explicitly on corporate ownership instead of geolocation.

Companies have already responded to the new policy, with Chinese startup Zhipu releasing a Claude-to-GLM-4.5 migration toolkit featuring "plug-and-play" switching, large context support, about 20 million free tokens, and higher throughput capabilities. Alibaba also pushed migration to Qwen-plus, bundling generous token allowances and aggressive pricing, as it did following earlier OpenAI restrictions.