DeepSeek misused in stock market manipulation scheme
Traders reportedly used DeepSeek to generate false information about A-share company partnerships by posting fake claims on other platforms, feeding them to the AI model, and circulating its misleading responses to manipulate retail investors. At least 12 listed companies issued denials, and one company saw a 46% stock surge based on such false AI-generated claims.
What happened
People posted false claims on other platforms about A-share listed companies partnering with DeepSeek, then fed these false claims to DeepSeek to generate misleading responses, and circulated these AI-generated answers online to further feed back into DeepSeek.
Retail investors screenshotted DeepSeek's affirmative responses to questions about company partnerships and shared them in stock forums as reliable information, leading other investors to buy shares based on the AI-generated false claims.
Reported impact
Twelve A-share listed companies, including Runhui Shares (润建股份), Sichuan Jinding (四川金頂), Hexin New Materials (禾盛新材), and others, published clarification announcements denying partnerships with DeepSeek. Sichuan Jinding rose 46.41% cumulatively over four consecutive trading days after 17 February, compared to Shanghai Composite Index growth of 0.12% in the same period.
Index analysis
The scheme exploits DeepSeek's tendency to generate affirmative responses to queries about partnerships and creates a feedback loop where false information enters the model's training context, potentially making the model more likely to generate similar false claims in response to future queries.
Governance controls
DeepSeek's user agreement states that AI model responses may contain errors or omissions and that outputs should not be treated as professional advice, but experts noted that once a model is partially polluted with false data, remediation is difficult.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.