A 67-year-old farmer in Chuzhou, China, lost nearly 25 acres of sesame seedlings after following an herbicide recipe generated by an AI assistant. The farmer had trusted the tool for months after it provided useful advice, but this time the AI's recommendation proved catastrophic.
AI's Fatal Recommendation
The AI generated a plan it called "Hundred Acres of Sesame Grass Control + Pest Control." It recommended using high-efficiency flupyrimethalin and flusulfasulfaether to kill weeds, mixed with thiamethoxazine and methyl salt. The farmer followed the instructions exactly, trusting the tool after months of successful interactions.
Agricultural experts later explained that flusulfasulfaether is designed for broadleaf weeds in soybean fields. The USDA categorizes sesame as a broadleaf plant, meaning the herbicide would kill the crop as effectively as the weeds. The chemical is also meant for targeted spraying, not whole-field application.
Immediate Consequences
The next day, the farmer saw both weeds and seedlings dying. "If you spray it, the next day the seedlings won't survive," Wu said in a video interview. "Both the grass and the seedlings will die, and the seedlings will die even faster."
When he asked the AI why the crop failed, it suggested the flusulfasulfaether might be the cause. The AI's chat page included a reminder that "AI generation may be incorrect, please verify," but the farmer did not see it or ignored it.
Why This Matters
This incident highlights a critical flaw in relying on AI for specialized knowledge. While large language models can generate plausible-sounding advice, they lack the deep domain expertise needed for agriculture, medicine or law. The farmer's trust, built over months of correct advice, led to a single catastrophic error. As AI tools become more accessible, users must understand that these systems are prediction engines, not experts. The real-world consequences of blind trust can be devastating, as this case shows.
Broader Lessons for AI Users
This is not an isolated case. AI has been known to give bad advice, from deleting user files to clearing inboxes. The key lesson is that AI outputs should always be verified, especially when they involve physical actions with irreversible results. The farmer's experience underscores the need for clear warnings and user education about AI limitations.



