China Warns Of AI Risks Citing Anthropic Claude Code
China’s industry regulator has issued a security warning about potential “backdoor” risks tied to Anthropic’s AI coding tool, Claude Code, according to multiple recent reports. The alert raises concerns about how the software could be used in ways that create or expose hidden access paths in systems.
The warning was reported by Reuters and described in coverage by CNBC, with China Daily and Global Times also carrying accounts of the regulator’s action. In that reporting, the regulator flagged the risk that an AI coding product could introduce security vulnerabilities that might be exploited, framing the issue as a “backdoor” threat.
The development centers on Claude Code, an AI-assisted coding tool from U.S. company Anthropic. The warning comes as governments and watchdogs globally intensify scrutiny of AI tools used in software development, particularly as organizations embed them deeper into programming workflows and critical systems.
The regulator’s action matters because AI coding assistants are increasingly used to generate, modify, and review code. Any security issue—whether in the tool itself or in the code it helps produce—can be replicated quickly and at scale across products and networks. A formal warning from a national regulator can also influence how companies inside that market evaluate, restrict, or replace widely used software.
The reports land amid a broader set of security and policy signals around AI. The Financial Times recently reported that top banking watchdogs issued a stark warning over AI-driven cyber attacks, reflecting growing concern that the same technology improving productivity can also accelerate malicious activity. Separately, Decrypt reported that Anthropic removed a hidden Claude Code tracker after researchers raised privacy concerns, adding to the attention on how the tool operates and what data may be collected or transmitted.
Corporate responses are also emerging. Benzinga reported that Alibaba banned Anthropic’s Claude for employees, citing security risks, and directed staff to use Qoder instead. That report suggests large organizations in China may be taking precautionary steps as official warnings circulate and internal risk teams reassess exposure.
What happens next will depend on any follow-up actions from Chinese authorities and how companies adjust their policies on AI coding tools. Organizations that rely on Claude Code may increase internal testing, tighten approval processes, or shift to alternative products as they evaluate security guidance and compliance expectations. Anthropic’s response, and any technical clarifications about the reported “backdoor” concerns, will be closely watched by enterprise customers and security researchers.
The warning underscores a fast-hardening reality for AI software: as tools like Claude Code move from experimentation into core development pipelines, security and oversight are becoming as central to adoption as performance.
