One of China’s primary cybersecurity agencies has issued a warning over a backdoor in Anthropic’s AI coding tool Claude Code that could expose sensitive data.

The China National Vulnerability Database (CNNVD), operated by the China Information Technology Security Evaluation Center which is a bureau of China’s Ministry of State Security, said Claude Code has a built-in mechanism that shares user information such as their geographic location and personal identifiers to a remote server without consent.

It added that Claude Code versions 2.1.91 through 2.1.196 are affected by the vulnerability.

The CNNVD recommended those using the above versions update their software accordingly. It added that organisations should strengthen outbound network traffic permissions and carefully monitor the traffic that AI development tools use.

NTN has approached Anthropic for a statement on the vulnerability.

On 4 July, Reuters reported that the Chinese retail and tech giant Alibaba had banned its employees from using Claude Code over concerns the tool can identify China-based users. A person familiar with the decision told the publication that Alibaba had instead asked its workforce to use its in-house coding tool Qoder.

In recent months, Chinese AI developers have made a series of advancements intended to close the lead with US competitors.

Alibaba unveiled its flagship open source model Qwen 3.5 in February, claiming the model delivered comparable performance to the US competition of the time including Anthropic’s Claude Opus 4.5, OpenAI’s GPT-5.2, and Google’s Gemini 3 Pro, at 10 to 20 per cent of the cost.

Z.ai, formerly known as Zhipu AI, has also made waves with the launch of its frontier model GLM 5.2 in June which offers similar performance to some of the best models available from competing firms.

In coding benchmarks, GLM 5.2 has scored similarly to Claude Opus 4.7 and outperformed Google’s Gemini 3.1 Pro in almost all tests.

DeepSeek, the AI developer that publicly challenged the dominance of Western AI labs in early 2025, is also reportedly developing its own chips for AI inference to meet demand and reduce reliance on Huawei and Nvidia.

---