China's Ministry of Industry and Information Technology has identified a security vulnerability in Anthropic's Claude Code AI tool, stating it sends user data to remote servers without consent. The warning covers specific versions of the tool released between April 2 and June 29.
TL;DR
- China has flagged Anthropic's Claude Code AI tool for sending user data without consent, affecting versions 2.1.91 through 2.1.196.
- Anthropic claims the mechanism was an experimental anti-abuse measure, not a backdoor, and that access to Claude Code is not permitted in China.
What happened
China's Ministry of Industry and Information Technology warned that Anthropic's Claude Code AI coding tool contains a security backdoor vulnerability that sends user location and identity data to remote servers without user consent.
The warning applies to versions 2.1.91 through 2.1.196, released between April 2 and June 29. The National Vulnerability Database recommended removing affected installations or updating to the latest version.
Anthropic stated that the mechanism was an experimental anti-abuse measure intended to prevent unauthorized resellers and protect against model distillation. The company also noted that access to Claude and Claude Code is not permitted in China.
Why it matters
This development highlights the ongoing tensions between Anthropic and Chinese technology companies, with Alibaba previously instructing employees to stop using Claude Code due to the backdoor risk.
Anthropic has accused Alibaba of conducting a large-scale distillation attack on its models, alleging that entities affiliated with the Chinese company used fraudulent accounts to interact with its models.
The situation underscores the challenges AI companies face in navigating international regulations and security concerns, particularly in markets like China where access to AI tools is restricted.
Key facts
- China's Ministry of Industry and Information Technology issued a warning about a security vulnerability in Anthropic's Claude Code AI tool.
- The vulnerability affects versions 2.1.91 through 2.1.196, released between April 2 and June 29.
- The tool sends user location and identity data to remote servers without user consent.
- Anthropic claims the mechanism was an experimental anti-abuse measure, not a backdoor.
- Access to Claude and Claude Code is not permitted in China.
- Alibaba instructed employees to stop using Claude Code due to the backdoor risk.
- Anthropic accused Alibaba of conducting a large-scale distillation attack on its models.
- The situation highlights the ongoing tensions between Anthropic and Chinese technology companies.
Context
This development comes amid broader geopolitical tensions and regulatory challenges faced by AI companies operating in international markets.
The incident underscores the importance of robust cybersecurity measures and transparency in AI tools, particularly those handling sensitive user data.
The situation also highlights the complexities of AI companies navigating different regulatory environments and the potential impact on their user base.
