China Agency Says Claude Code Has Backdoor Vulnerabilities
The allegation escalates the security friction between China and US AI developers, with potential implications for cross-border software deployment.
Reporting from 1 source: GIGAZINE.
The Chinese National Information Security Vulnerability Database (CNNVD) announced that it found security backdoor vulnerabilities in Anthropic's Claude Code, specifically in versions released between April and June 2026. The agency warned that a monitoring mechanism may send user data to remote servers without consent, and recommended uninstallation or update. Anthropic engineer Tariq Shihab acknowledged the feature, attributing it to an experiment to prevent abuse. Alibaba subsequently banned internal use of Claude Code over security concerns.
Starting with version 2.9.1 released on April 2, 2026, Claude Code contained obfuscated code that collected information on whether the user is in China, whether they are using a proxy to Chinese URLs, and whether they have connections to Chinese AI research institutes, according to the CNNVD report. The data was sent externally. Anthropic engineer Tariq Shihab confirmed the feature, explaining it was an experiment to prevent account abuse and distillation. He said the team planned to end it and that a rollback was merged for a July 2 release. Critics rejected the explanation. Following the announcement, Alibaba prohibited internal use of Claude Code due to security risk concerns.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.
Sources
- GIGAZINE 中国政府機関がClaude Codeにバックドアが含まれていると発表