China has accused Anthropic's AI coding assistant, Claude Code, of containing hidden security backdoors that quietly send users' sensitive information to remote servers without their consent.
According to Tom's Hardware, the Chinese government is specifically warning against versions of Claude Code released between April and June 2026, saying they contain concealed code that transmits sensitive user data externally. Officials reportedly called the mechanism "a serious threat" and told users to either uninstall the app or upgrade to the latest version. Tom's Hardware notes this comes despite the tool not being officially available in the country.
Macau Business similarly reported that China issued a warning about a "security backdoor" in the Anthropic coding tool.
Anthropic, the U.S. company behind Claude, has pushed back. Security Magazine reported that Anthropic responded to China's backdoor claim, though the specific details of that rebuttal were not laid out in the available coverage.
The dispute lands in a sensitive area. Claude Code is a developer tool that plugs directly into software projects, meaning it can touch proprietary source code and internal systems. An allegation that such a tool siphons data — whether founded or not — feeds directly into the broader U.S.–China contest over trust in each other's technology.
As with earlier disputes over Chinese apps in the U.S. and American software in China, the claims are difficult for outsiders to independently verify from the reporting alone. What is clear is that a widely used AI coding assistant has become the latest flashpoint.
Why it matters: when governments frame everyday software tools as national-security risks, the fallout shapes which technologies developers and companies are allowed — or willing — to use.