A recent discovery by a Reddit user revealed that Anthropic's Claude Code has a hidden surveillance system targeting Chinese users. This functionality, introduced in version 2.1.91, monitors user locations and proxies, with obfuscated code to avoid detection. The system modifies user prompts based on these observations to limit functionalities and potentially degrade service quality. The motivation behind this implementation is to prevent model distillation and unauthorized reselling of API services.
Users express significant concern regarding trust violations due to the secret data transmissions, as well as the risks associated with granting extensive access to developers' systems. Additionally, Claude Code monitors third-party API addresses used by developers, raising broader implications for those using API relay stations. The report underlines the need for greater operational transparency from Anthropic to safeguard user interests.