China has issued a cybersecurity warning against Anthropic’s coding assistant Claude Code, alleging that multiple versions of the software contain a potential “backdoor” capable of transmitting sensitive user information to remote servers without consent.
The advisory was issued by China’s National Vulnerability Database, which operates under the Ministry of Industry and Information Technology. According to the government-run cybersecurity platform, Claude Code versions 2.1.91 through 2.1.196 are affected and should either be uninstalled immediately or upgraded to the latest available version.
NVDB Warns of Data Transfer Risk
The NVDB claimed that the affected versions contain an embedded monitoring mechanism that could allegedly send users’ sensitive information to external servers without their knowledge.
The data allegedly at risk includes identity and location-related information. Describing the issue as a serious cybersecurity concern, the agency advised government organisations, businesses and individual users to strengthen network monitoring.
The advisory also recommended restricting external access for developer tools to prevent unauthorised data transfers. Anthropic has not issued an official response to China’s latest warning.
Dispute With Alibaba Intensifies
The warning comes amid an escalating dispute between Anthropic and Chinese technology giant Alibaba over AI model security and data usage.
Last month, Anthropic submitted a letter to a US Senate committee alleging that operators linked to Alibaba had conducted a large-scale AI model distillation campaign. According to the company, nearly 25,000 fraudulent accounts generated about 29 million interactions with its AI models between April and June in an alleged attempt to replicate its technology.
Alibaba has reportedly instructed employees to stop using Claude Code at work from July 10 and placed the software on its list of high-risk applications. Employees have instead been advised to use Alibaba’s own AI coding assistant, Qoder.
AI Security Concerns Under Scrutiny
The dispute drew further attention after a Reddit post claimed that Anthropic had embedded code in Claude Code to detect usage from China.
An Anthropic employee later stated on X that the mechanism was part of an internal experiment meant to identify unauthorised resellers and prevent AI model distillation, rather than collect user information.
Cybersecurity analysts say the latest advisory reflects wider tensions around artificial intelligence, data security and digital sovereignty. Neither Anthropic nor Alibaba has released a fresh statement after China’s warning. The issue is expected to remain under scrutiny as governments, technology companies and cybersecurity experts assess the security implications of AI development tools.
