A man in Zhuji, Zhejiang, was rushed to the intensive care unit (ICU) after eating wild mushrooms that an AI app had incorrectly flagged as safe, leading to acute kidney failure and rhabdomyolysis. The incident has reignited public debate over whether AI-driven mushroom identification tools can be trusted with human lives.
A single meal of "AI-certified" wild mushrooms landed him in critical condition. According to local reports, the patient, identified only as Lao Wang, collected fresh wild mushrooms near his home and, as a precaution, photographed them for an AI app to verify their safety. After receiving a "non-toxic" confirmation, he cooked and ate them at noon. Hours later, he experienced five bouts of watery diarrhea, severe pain in both arms, and overwhelming fatigue. Rushed to the hospital, his blood tests showed creatine kinase levels at 28,771 U/L, nearly 100 times the normal range, accompanied by acidosis and hyperlactatemia, confirming acute kidney failure and rhabdomyolysis. He was immediately admitted to the ICU, where treatment focused on toxin elimination, supportive care, and organ protection, eventually leading to his full recovery and discharge.
In a test of AI identification accuracy, five different wild mushroom species were presented to an AI tool via video call. The AI identified them as Yellow-White Athena Mycena, White Coprinellus, Hard-Stemmed Marasmius, Large Green Spore Mushroom, and Large-Cap Marasmius, with some flagged as toxic or resembling lethal species. However, an expert from the Chinese Mycological Society and Zhejiang University, Lin Wenfei, identified the actual species as Yellow-White Athena Mycena, Hard-Stemmed Marasmius, Hard-Stemmed Marasmius, Large Green Spore Mushroom, and Gymnopus. The AI was wrong on two of the five, mistaking a Hard-Stemmed Marasmius for a White Coprinellus and a Gymnopus for a Large-Cap Marasmius. Even more concerning, when a second phone was used on the same mushroom, results varied further, with one identification mismatching the expert's verdict entirely.
This is not an isolated case. In June, a senior executive at ByteDance publicly addressed an incident where their Doubao AI misidentified a toxic mushroom as edible, potentially exposing a user to a lethal species. The executive warned that identifying wild mushrooms solely from images carries extreme risk, as similar-looking toxic species cannot be fully ruled out.
AI recognition technology relies on computer vision and machine learning, but it struggles with variations in lighting, angle, background, and scale. Mushrooms, unlike plants with distinct roots, stems, leaves, flowers, and fruits, offer limited identifying features, and toxic species often closely resemble edible ones, with no absolute patterns in color or texture. Even professional mycological guides cannot guarantee perfect accuracy. A researcher from the Chinese Center for Disease Control and Prevention noted that AI mushroom identification remains immature, with no dedicated software developed domestically. Studies show that even the best AI tools fail nearly 15% of the time in real-world tests. Major AI platforms themselves now caution that they cannot precisely distinguish toxic mushrooms and should not be used as a basis for consumption.
The stakes are deadly. CDC data from 2025 recorded 828 mushroom poisoning incidents nationwide, affecting 2,165 people and causing 13 deaths. Symptoms range from gastrointestinal distress to severe liver and kidney failure, with no specific antidote available. Highly toxic species containing amatoxins account for over 90% of fatalities, and ingesting just 50 grams can kill a healthy adult. Some lethal mushrooms have a fatality rate exceeding 80%. The deadly Amanita species often has a 6- to 12-hour latency period, initially mimicking common gastroenteritis, which can delay life-saving treatment beyond the critical 48-hour window.
Experts emphasize that unfamiliar wild mushrooms must never be collected or consumed, regardless of AI confirmation. Field surveys in urban parks and residential areas have found that over half of the mushroom species sampled were toxic, and these dangerous fungi thrive in accessible locations like lawns and green belts, especially during the rainy season from June to September. For known edible varieties, experts advise boiling for at least 20 minutes before consumption. They also highlight a simple rule: mushrooms with a cap, ring, and volva, resembling a figure wearing a hat, apron, and boots, should never be eaten, no matter how similar they look to safe species.
As AI technology advances rapidly, its limitations in life-critical applications cannot be overlooked. AI identification should serve only as a reference, never as a license to eat. The experts' warning is clear: encountering wild mushrooms means do not touch, do not pick, and do not eat, even with AI approval. An AI can be retrained after a mistake, but a human life cannot be replaced.