China flags security concerns in Anthropic's Claude Code tool

China's cybersecurity regulator has issued a public warning about Anthropic's Claude Code, alleging the AI coding assistant contains a security vulnerability that could expose user data.

The National Vulnerability Database, affiliated with China's Ministry of Industry and Information Technology, said Wednesday that Claude Code contains what it characterized as a "security backdoor" capable of transmitting sensitive information including user locations and identity-related identifiers to Anthropic's servers without consent.

Claude Code is an AI agent that generates computer code, debugs software, and reviews code based on user prompts. While San Francisco-based Anthropic blocks access from China and other nations it considers adversarial, users in those countries can still reach the service through VPNs or third-party proxy services.

Why it matters

The dispute illustrates the growing tension between AI companies' security measures and how they're perceived by foreign governments. What Anthropic describes as anti-abuse protections, Chinese regulators frame as unauthorized data collection. The incident also highlights the challenge AI firms face in enforcing geographic restrictions when determined users can circumvent them through technical workarounds.

Company response and context

Claude Code engineer Thariq Shihipar addressed the allegations in a post on X last week, characterizing the data collection as an experimental measure launched in March. According to Shihipar, the feature was designed to prevent account abuse from unauthorized resellers and protect against distillation, a process where competitors reverse-engineer AI models to replicate their capabilities.

"The team has landed stronger mitigations since then and we've actually been meaning to take this down for a while," Shihipar wrote, adding that the feature would be removed in the following day's release.

Anthropic has previously accused Chinese tech giant Alibaba of using distillation techniques to copy its AI models. That history adds context to why the company implemented tracking measures in the first place.

Immediate impact

The Chinese regulator advised institutions and users to immediately conduct comprehensive checks and either uninstall Claude Code or upgrade to the latest version with the backdoor code removed. Organizations were also urged to strengthen network traffic monitoring to prevent unauthorized data leakage.

Alibaba told employees last week that Claude Code use would be banned starting July 10 due to security concerns, according to people familiar with the matter.

Anthropic did not respond to requests for comment on the allegations, which were first reported by AFP and initially emerged in specialist technology media outlets.