Approximately one in 12 British and American employees use Chinese generative AI (GenAI) tools, exposing their organizations to security, privacy and compliance risks, according to Harmonic Security.
The security vendor carried out a 30-day analysis across 14,000 US and UK-based end users to better understand the risks of using Chinese GenAI tools like DeepSeek, Moonshot Kimi, Manus, Baidu Chat and Qwen.
Although these tools offer powerful capabilities, usually for free, they are considered too risky by most Western security leaders. That’s because any data shared with the model will be stored on servers in China, where the authorities could access it.
Most Chinese GenAI providers are also vague about how they might use this data, creating further legal liabilities.
Read more on GenAI: Chinese GenAI Startup DeepSeek Sparks Global Privacy Debate
Harmonic Security’s study found that the 1059 users who used Chinese GenAI tools uploaded 17MB of content. The vendor claimed to have detected 535 incidents of sensitive data exposure, most of which occurred via DeepSeek – accounting for 85% of the total.
Most exposed content was software related: 33% involved code, proprietary logic or access credentials. Other sensitive categories included M&A documents (18%), personally identifiable information (18%) and financial data (14%).
“All data submitted to these platforms should be considered property of the Chinese Communist Party given a total lack of transparency around data retention, input reuse, and model training policies, exposing organizations to potentially serious legal and compliance liabilities,” warned Alastair Paterson, CEO of Harmonic Security.
“Blocking alone is rarely effective and often misaligned with business priorities. Even in companies willing to take a hardline stance, users frequently circumvent controls. A more effective approach is to focus on education and train employees on the risks of using unsanctioned GenAI tools, especially Chinese-hosted platforms.”
Organizations should also provide more secure alternatives and enforce policies preventing sensitive data from being uploaded, he added.
Researchers have also found DeepSeek is vulnerable to jailbreaking, hallucinations, generating insecure code, and that its Android app is riddled with security issues.
Additionally, the firm accidentally leaked an online database back in January that contained sensitive information including chat histories.