China's National Vulnerability Database Issues Security Alert for Anthropic's Claude Code Models

China's National Vulnerability Database (NVDB), a cybersecurity entity linked to the Chinese government, has raised significant concerns regarding certain iterations of Anthropic's Claude Code AI models. The NVDB has advised users against utilizing these versions without immediate updates, citing serious security vulnerabilities.

According to a report from Reuters, the NVDB claims that specific versions of Claude Code contain an embedded monitoring feature. This mechanism is alleged to transmit confidential user data, such as precise geographical location and personal identification details, to external servers without explicit user consent. These allegations were disseminated through the agency's official WeChat account.

The identified security flaw purportedly impacts Claude Code versions ranging from 2.1.91 to 2.1.196. It is important to note that version 2.1.196 was subsequently superseded by version 2.1.200 on July 3rd. The NVDB has urged all users of the affected versions to either completely remove the software from their systems or promptly apply the latest available update to safeguard their data.

In response to these revelations, the NVDB has also recommended that users implement more stringent controls and enhance monitoring protocols for any data provided to Claude Code. Anthropic has yet to issue a public statement regarding these claims.

This is not the first instance this month where Anthropic's coding tool has faced scrutiny in China. Earlier, Alibaba reportedly prohibited its employees from using Claude in their professional capacity. This decision stemmed from concerns that the AI tool could potentially identify Chinese users based on the information it processed.

The intricate relationship between Anthropic and Alibaba extends further, with Anthropic previously accusing the Chinese e-commerce giant of leveraging its intellectual property to train Alibaba's own AI models. Anthropic alleged that between late April and early June, Alibaba orchestrated over 28 million interactions with Claude through nearly 25,000 fraudulent accounts, ostensibly to harvest data. Notably, Anthropic has a policy of restricting access to Chinese companies, a limitation that these companies have reportedly circumvented through overseas subsidiaries and the use of Virtual Private Networks (VPNs). The Financial Times reports that Anthropic is actively seeking ways to counteract these workarounds in the future.

The current climate suggests a growing tension in the relationship between Anthropic and Chinese entities, potentially leading to a shift towards domestically developed AI solutions within China.