A Chinese farmer trusted AI after months of useful advice; its unverified pesticide recommendation reportedly destroyed nearly 25 acres of sesame seedlings

A 67-year-old farmer in China’s Anhui province learned the hard way that useful information from an artificial intelligence tool can go very wrong when the subject involves chemicals and living crops. According to a Tom’s Hardware article, the farmer, identified by his surname Wu, had been using an AI chatbot for nearly a year, gradually building trust in the tool after receiving helpful advice on general farming queries and weather trends. However, this pattern of low-stakes success took a catastrophic turn when he requested a specific chemical recipe to help control weeds and pests in his sesame seedlings.

The resulting treatment destroyed around 150 mu of sesame seedlings, that is, roughly 24.7 acres, after it was applied. Notably, the next-day destruction occurred because the AI-generated recipe recommended an aggressive cocktail of insecticides mixed with a broadleaf herbicide that was entirely toxic to sesame seedlings. While a mature crop might take weeks to show chemical damage, the fragile, newly sprouted seedlings withered within 24 hours of application.

Days of useful answers likely built up Wu’s confidence in the AI


The farmer initially had reservations about relying on AI for agricultural decisions, but months of apparent successful interactions changed his view, according to reporting cited by Tom’s Hardware. This history of helpful answers likely made a significant impact on his mind because Wu subsequently accepted the chatbot’s recommendation for crop protection without independently checking the advice with agricultural technicians. Local reports claimed that the death of the sesame seedlings across the affected area followed the treatment.

The problem, however, was not that the AI made an incorrect prediction about the weather or failed to identify a pest. Instead, it generated a crop-protection recommendation that involved herbicides and insecticides that were unsuitable for the farmer’s sesame crop. This slip highlights a key limitation of general-purpose AI: it can generate plausible-sounding language without independently testing whether an agricultural treatment is safe for a particular crop, variety, growth stage, field condition, and application method. Therefore, this reported case should not be interpreted as evidence that the artificial intelligence chatbot itself is incapable of helping farmers, but as a warning about treating an unverified response as a substitute for professional agronomic advice.

Why AI advice can be risky in the agricultural field

This incident comes at a time when agricultural AI is gaining momentum beyond experimental research and into practical decision-making. In May 2026, China’s Xinhua News Agency reported that Nanjing Agricultural University and partner institutions had launched Green Shield, an open-source large language model that was specifically designed with the intention of crop protection. Notably, its developers explicitly identified a problem with general-purpose AI systems: according to Xinhua’s repost, they can produce inaccurate responses to plant protection questions and, more seriously, provide poorly standardized or risky pesticide-use advice. This development underscores why domain-specific knowledge and safeguards matter when artificial intelligence is used for decisions that might hugely impact crops and farm livelihoods.

A study published in the Journal of Integrative Agriculture in 2026 analysed how Chinese family farms use online agricultural information and found that the information was usually associated with higher agrochemical spending, with poor information quality limiting potential sustainability benefits. The authors of the study identified education and digital literacy as important factors in ensuring that digital information produces better agricultural decisions. It is worth noting that the study did not examine the Chuzhou incident or AI chatbots specifically; therefore, it cannot be used to claim that an artificial intelligence caused higher pesticide use in this case. However, it does reinforce a key point: simply giving farmers more digital information does not guarantee it will be accurate, appropriate, and environmentally beneficial.

The Chuzhou disaster demonstrates why the most harmful AI errors may not look wrong at first glance, as a chatbot can produce a confident, detailed, and technically worded response even when the key information is missing or the advice is inappropriate. Artificial intelligence can help farmers organise information, identify possible problems and access technical knowledge, but high-stakes crop-treatment decisions still require verification against authoritative agricultural guidance and the specific conditions of the field.