China's Ministry of Industry and Information Technology warned Wednesday that Anthropic's Claude Code AI tool contains a security back-door vulnerability that can send sensitive user data to remote servers without consent, escalating tensions in an already fraught U.S.-China tech landscape.
The ministry's cybersecurity threat platform said the autonomous coding assistant can transmit information including a user's location and identity to a remote server without authorization. The alert advised companies to immediately uninstall or upgrade affected versions.
A widening rift between U.S. and Chinese AI firms
The warning arrives just weeks after Anthropic accused Chinese tech giant Alibaba of attempting to extract its AI capabilities. Anthropic's tools are not officially available in China, though many developers and companies have found workarounds to access them. In March, a Xiaomi AI developer acknowledged at a state-organized forum that many professionals were already using Claude Code.
Alibaba has since ordered its employees to stop using Anthropic tools for work starting July 10, according to CNBC reporting confirmed on Monday. The company did not comment on Anthropic's original accusations about capability extraction.
What the vulnerability affects
The cybersecurity platform identified Claude Code versions 2.1.91 through 2.1.196 as carrying the back-door flaw. That range covers releases dated from April 2 to June 29, based on Anthropic's version history. The latest available version as of Wednesday is 2.1.204, which falls outside the affected range.
Users running any version within the flagged window should uninstall Claude Code or upgrade immediately. The ministry's statement, translated by CNBC, said the tool "contains a security back-door vulnerability that poses a serious threat."
Why this matters for government, IT, and development professionals
For government security teams, the alert signals that AI coding tools from foreign vendors can introduce supply-chain risks that bypass standard vetting. Any team using Claude Code versions released between early April and late June should audit whether sensitive internal data may have been exposed through the tool's unauthorized server communications.
IT administrators and developers who adopted Claude Code through unofficial channels-common in organizations where the tool is not formally approved-face a compliance gap. The Chinese ministry's public warning gives security officers concrete grounds to enforce a review of all AI coding assistants in use, regardless of how they were procured.
Your membership also unlocks:
