Regarding the recent online controversy about "Doubao AI misidentifying mushrooms leading to user poisoning," the Doubao team, a multimodal large model from Volcano Engine, has officially contacted the affected user and verified and analyzed the entire incident process.

The feedback showed that the user previously used Doubao to take a photo of wild mushrooms picked in the neighborhood. Although the system identified them as "chicken leg mushrooms," it also clearly provided multiple high-risk warnings in the chat window. The system pointed out at that time that this appearance could easily be confused with the highly toxic "Giant Puffball" and that consuming it could lead to severe gastroenteritis, and it clearly issued "strongly advised not to eat" and "it is not possible to 100% exclude the possibility of toxic similar species based only on the image" disclaimers and safety tips.

QQ20260605-145844.jpg

For such complex and high-risk scenarios involving personal safety, botanists and industry experts point out that the current visual recognition of multimodal AI is still in the technical development stage. Even professional scholars may have identification errors when facing fungi in unfamiliar areas. Some highly toxic mushrooms have extremely high mortality rates and cannot be experimented with. The Doubao team responded that the large model is continuously optimizing and improving the accuracy of identification.

This incident has once again sparked deep thinking in the industry about the safety boundaries of large model applications. Industry experts remind the public that AI outputs related to life safety and health are only for reference, and users must consult and verify through multiple authoritative channels. At the same time, wild fungi or fruits in urban community green belts are prone to residual pesticides and other unknown pollutants. Even if the species itself is non-toxic, there is a very high food safety risk, and the public should resolutely avoid eating them.