Anthropic closes China relay loopholes

Anthropic is moving to block Chinese companies from accessing Claude through overseas subsidiaries, VPNs, and third-party relay services, the Financial Times reported on July 3. Companies including Ant Financial had provided employees corporate Claude accounts through Singapore-registered entities, while ByteDance reimbursed engineers for personal accounts accessed via VPN.

From April 2026, Anthropic extended its restrictions to cover majority-owned subsidiaries of restricted entities and began requiring government-issued photo ID and a live selfie from accounts flagged as suspicious. The company monitors for computer time zone indicators as part of its detection approach and has moved to shut down commercial “transfer station” services that relay requests through overseas accounts.

Some covert detection measures were rolled back following public pushback, though Anthropic has not disclosed which ones.

The context

The relay crackdown connects directly to Anthropic’s accusations of distillation by Alibaba: roughly 25,000 fraudulent accounts generated 28.8 million Claude interactions to systematically extract Claude’s capabilities. The Alibaba ban on Claude Code, which surfaced when telemetry code in the tool was discovered, is a separate downstream consequence of the same underlying conflict.

For operators

If your Claude deployments involve accounts linked to China-based entities, or if your users access through VPNs or offshore proxies, expect potential verification challenges. The monitoring covers both bad actors and legitimate users routing through privacy tools or regional infrastructure.

Operators serving customers in or near restricted jurisdictions should review their access architecture. The Claude apps gateway, released July 2, routes inference through your own Bedrock or Google Cloud account and may reduce exposure to account-level monitoring.