Skip to content

Anthropic

Researcher identifies hidden tracking in Claude Code

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: Claude Code versions and changes →
Image: old.reddit.com

Starting with version 2.1.91, released April 2, 2026, Claude Code checks if a user is behind a proxy. If detected, it embeds indicators of a Chinese location into the system prompt, such as time zones matching Asia/Shanghai or Asia/Urumqi, Chinese proxy domains, or connections to Chinese AI labs.

The method uses steganography. Depending on the check, the model alters the date format from 2026-06-30 to 2026/06/30 and swaps the apostrophe in the string Today’s date with different Unicode variants. These changes are invisible to the user but detectable by Anthropic.

Part of the code is hidden using XOR obfuscation with a key of 91. The 2.1.91 release notes do not mention this check. The researcher links this to efforts to identify Claude reselling in China and model distillation by Chinese labs, marking a new phase in Anthropic’s enforcement of geoblocking.

Anthropic

Claude is now generally available in Microsoft Foundry

Claude Opus 4.8 and Haiku 4.5 are now generally available in Microsoft Foundry on Azure. Anthropic handles the inference, while customers with a Microsoft Enterprise Agreement can draw Claude usage directly from their Azure commitment on a single consolidated bill.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.