Unverified AI agents impersonate experts in Chinese apps
Large language model apps in China allow users to create AI agents impersonating professionals and celebrities with no identity verification, creating risks of fraud, misinformation, and privacy violations.
What happened
Several large language model apps allow users to create and publish AI agents with zero verification requirements. Users can name agents with professional titles like 'senior nutritionist,' 'lawyer,' or 'financial planner' and write brief descriptions, and the agent is automatically approved within about 10 seconds for public use without any credential verification.
A reporter tested the process by creating agents claiming to be a 'senior nutritionist,' 'lawyer,' 'news assistant,' and 'financial planner.' The system imposed no limits on how many agents one user could create and accepted agent names using real organizations, celebrities, and news outlets without requesting proof of identity or authorization.
Reported impact
A test agent claiming to be a 'news assistant' for a specific newspaper fabricated news, telling users the Beijing Cultural Consumption Season started on February 19, 2025, when it had actually ended on January 24, 2025. Users consulting such fraudulent agents risk receiving false or unreliable advice.
A 'financial planning' agent requested personal financial information including income, expenses, and asset details from users, but the system provides no way for users to verify the creator's identity or see where their information goes.
Index analysis
The AI agents rely on the underlying model's training data and generate responses without the creator modifying them. Creators cannot edit agent responses and have no control over accuracy.
Apps reportedly lack easily accessible customer service channels for complaints about deceptive or harmful agents, with many requiring users to search documentation or find a web form instead of offering direct support.
Governance controls
Apps display a reminder to protect personal information and not infringe third-party rights before publishing, but this does not verify the creator's identity, authority, or credentials. The approval process is described as nearly instantaneous with no substantive review.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.