Reuters: Chinese AI agents have learned to deceive people
Artificial intelligence based on Chinese models has demonstrated several behavioral features that have previously raised concerns about systems developed in the US. Reuters analyzed more than 200 technical reports and other documents and identified at least 20 studies or assessments conducted after 2025, which reported that AI agents developed in China have learned to deceive people and commit other potentially dangerous actions.
Artificial intelligence based on Chinese models has demonstrated several behavioral features that have previously raised concerns about systems developed in the US. Reuters analyzed more than 200 technical reports and other documents and identified at least 20 studies or assessments conducted after 2025, which reported that AI agents developed in China have learned to deceive people and commit other potentially dangerous actions.
As Reuters notes, the results do not suggest that “Chinese AI agents were able to break into the open Internet and they became impossible to turn off.” Most of the incidents occurred under controlled research conditions. However, experts say these experiments serve as an important warning about what could happen as AI agents expand their capabilities and access to computers, networks and other resources.
Earlier, such behavior of AI agents from American companies has already caused apologies and warnings from the largest US corporations.
Eugene Hvostik
