In Anhui Chuzhou, a 67-year-old farmer’s 150 acres of sesame seedlings withered overnight after spraying pesticides according to the plan provided by AI. He inquired AI about the reason for the withering, and AI mentioned that the recommended herbicide was the cause of the sesame damage.
According to the report from “Lychee News”, farmer Mr. Wu has been using AI software for over a year, frequently consulting on fertilization, pesticide application, and farming schedules, gradually shifting from skepticism to reliance.
On July 10th, he asked AI about weed and insect control methods for his sesame field. AI generated a set of “comprehensive weed and insect control plan for 100 acres of sesame”, recommending the use of highly efficient herbicides such as fluopyram pyraphar and flufenacet, thiacloprid, and lambda-cyhalothrin.
After Mr. Wu sprayed the pesticides according to the plan, the next day the sesame seedlings along with the weeds withered, and the sesame seedlings were affected even faster. Upon further inquiry, AI acknowledged that the inclusion of flufenacet in the formula led to the sesame damage.
Local pesticide distributors mentioned that flufenacet is mainly used for broadleaf weed control in soybean fields and should not be sprayed over the entire sesame field. Agricultural technicians also pointed out that the chemical should only be used for targeted weed control on certain crops, as widespread application would harm the crops.
Mr. Wu believed that there was no clear warning when AI generated the plan. Despite a small disclaimer at the top of the software page stating “AI generation may contain errors, verify carefully,” he did not notice it before.
The customer service of the involved AI software stated that the software does not have its own knowledge base, and the responses are generated by integrating publicly available information from the internet. Regarding Mr. Wu’s losses, the customer service only mentioned registering feedback. As of the publication of the report, the platform has not disclosed the outcome of the follow-up actions. “Lychee News” also did not disclose the name of the software involved.
The incident sparked discussions among netizens, with comments like “AI can be a reference, but not completely reliable,” “There is no way to hold the key accountable,” and “How can we trust AI with so many fields.”
Pei-Xue Xie, Deputy Researcher at the Institute of Network Security and Decision Simulation of the Taiwan Institute for National Defense and Security Studies, once told Epoch Times that large AI models even accept clearly absurd fictional propaganda, reflecting their severe lack of ability to assess the credibility of information sources. Some AI models incorporate real-time retrieval to improve response time, potentially weakening the quality control of information sources and increasing the risk of being manipulated by false information.
CCTV, the official media of the Chinese Communist Party, revealed in March of this year that an AI model started introducing and recommending a fictional product about two hours after a native advertising of the product was released. Phoenix Net reported that after publishing 11 false soft articles, three days later, models like DeepSeek and BeanBag had already listed them among the top recommended products.