Anthropic has defended the software checks in Claude Code that identify users accessing the AI coding tool from China, following claims of a 'security backdoor.' The checks are part of the company's anti-abuse measures, according to a report by the South China Morning Post.
TL;DR
- Anthropic's Claude Code includes checks to identify users accessing the tool from China, sparking claims of a 'security backdoor.'
- The checks are part of Anthropic's anti-abuse measures, targeting unauthorized access and model distillation.
- The dispute highlights the ongoing U.S.-China tensions over AI model access and distillation practices.
What happened
Anthropic has defended the software checks in Claude Code that can identify users accessing the AI coding tool from China. These checks were highlighted by China's National Vulnerability Database, which claimed that the tool contained a 'security backdoor.' According to the South China Morning Post, Anthropic stated that the code was an anti-abuse measure aimed at detecting unauthorized access from China and other unsupported regions.
China's National Vulnerability Database warned that Claude Code versions 2.1.91 to 2.1.196, released from April through June, could transmit sensitive information, including user location and identity-related data, back to Anthropic servers without user consent. The agency advised users to uninstall affected versions or upgrade to a newer release.
Anthropic clarified to AFP that the mechanism checks a device's time zone and whether a request was routed through an unsupported or banned region or to AI labs suspected of using Claude outputs to train their own models, a practice known as distillation. The company stated that China-based users were never authorized to use Claude Code.
Why it matters
This dispute comes at a time when U.S. AI companies are tightening controls over who can access their most capable models. Anthropic has accused several Chinese AI developers, including DeepSeek, Moonshot AI, MiniMax, and Alibaba Group, of using Claude for large-scale distillation attacks intended to improve their own models.
The controversy has intensified broader U.S.-China tensions over whether Chinese AI labs are using U.S. frontier models to improve their own systems. Last week, Chinese firm Alibaba banned employees from using Claude Code starting July 10 after researchers learned the code checked for China-related indicators.
Anthropic's actions highlight the growing importance of protecting AI models from unauthorized access and distillation, which can undermine the competitive advantage of AI companies. For developers and startups, this underscores the need for robust security measures to prevent model misuse and unauthorized access.
Key facts
- Anthropic's Claude Code includes checks to identify users accessing the tool from China, sparking claims of a 'security backdoor.'
- China's National Vulnerability Database warned about Claude Code versions 2.1.91 to 2.1.196, advising users to uninstall or upgrade.
- Anthropic stated that the checks are part of its anti-abuse measures, targeting unauthorized access and model distillation.
- The dispute involves claims that Claude Code could transmit sensitive information back to Anthropic servers without user consent.
- Anthropic accused several Chinese AI developers of using Claude for large-scale distillation attacks.
- Alibaba banned employees from using Claude Code starting July 10, adding it to its list of high-risk software.
- Anthropic sent a letter to two U.S. senators accusing Alibaba of carrying out 'the largest known distillation attack on Anthropic to date.'
- The letter stated that Alibaba and its Qwen AI lab generated 28.8 million exchanges with Claude through roughly 25,000 fraudulent accounts between April 22 and June 5.
Context
This controversy is part of a broader trend of U.S. AI companies tightening controls over model access and preventing unauthorized use. The dispute highlights the ongoing tensions between the U.S. and China over AI technology and the protection of intellectual property.
For AI/ML developers, this underscores the importance of implementing robust security measures to prevent model misuse and unauthorized access. Startup founders and tech investors should be aware of the potential risks and the need for proactive measures to protect AI models.
The competitive angle is clear: protecting AI models from distillation and unauthorized access is crucial for maintaining a competitive edge in the rapidly evolving AI landscape. The dispute also highlights the need for clear communication and transparency in addressing security concerns and potential vulnerabilities.
