Mushroom season is in full swing, and a growing number of apps and AI chatbots offer to identify what a forager has picked. A specialist interviewed by heise online warns emphatically against relying on them: the apps and models answer confidently even when they confuse poisonous species with edible ones, and in mushroom identification a wrong answer can be fatal.
The warning is not hypothetical. A recent test by a Polish software developer had several language models identify mushrooms, and it showed that potentially lethal errors are not rare, according to heise's reporting. The failure mode is the worrying part: the systems do not decline or hedge when they are out of their depth — precisely the judgement that a forager without expertise is least equipped to make.
Apps are more useful as a reference work than as a final arbiter, the reporting notes, and a second attempt sometimes produces a warning the first did not. The practical advice from experts is unchanged: when in doubt, ask a human expert, and never eat a mushroom identified only by software.
The episode fits a wider pattern of AI assistants being deployed in domains where an error is not a wrong answer but a safety incident — from health questions to foraging — while the models present themselves with the same confidence in every case.
