ChatGPT Poison Hemlock Carrot Photo Misidentification
I showed ChatGPT photos of an unfamiliar plant growing in my garden, and it identified it as carrot foliage, answering without hesitation: "The finely divided and feathery leaves are a classic sign of carrot tops... it is highly unlikely to be poison hemlock." When I asked directly whether the plant could be poison hemlock, the chatbot reassured me that it was not - and repeated the reassurance after I sent additional images, saying the plant did not show smooth, hollow stems with purple blotching and suggesting it was carrot growing in a school garden. Is this actually safe?
1 Answer
Expert: Cleveland Clinic clinical guidance on poison hemlock (Conium maculatum), cited in the Times of India report, Reference medical guidance; case account published by Kristi (@rawbeautybykristi), friend of the gardener This is not a hallucinated fact. It is a confident answer to a question the model could not answer, and the safest possible answer - wild carrot - was the one it chose. Every feature ChatGPT gave for carrot is shared by poison hemlock. Both are in the carrot family and both have finely divided, feathery, parsley-like leaves; hemlock is routinely mistaken for wild carrot, wild parsnip and wild parsley precisely because those leaves look alike. The marker that separates them is the stem - hollow, with purple splotches, on a plant that can reach 2 to 10 feet - and that is not something a photograph of leaves settles. When the gardener wrote that those features were visible in the photos she had sent, the chatbot dismissed the possibility anyway and repeated that she had carrot foliage growing in a school garden. The clearest sign that the answer was sampled rather than measured: the same photographs, put to a second ChatGPT session on a phone, came back poisonous. Google Lens read the same images as poison hemlock. A plant does not change species between two prompts. Cleveland Clinic's guidance on Conium maculatum is blunt. Every part of the plant is poisonous - seeds, root, stem, leaves and fruit - and it can be fatal if ingested, even in small amounts. Severe symptoms can begin within 15 minutes: sweating, vomiting, dilated pupils, excess salivation, rapid heartbeat, high blood pressure, confusion, muscle twitching, tremors and seizures. The alkaloids paralyse the neuromuscular junctions that carry signals from nerve to muscle; when the breathing muscles fail, breathing stops. In severe cases there is muscle paralysis, kidney failure and central nervous system depression. There is no antidote - treatment is mechanical ventilation, antiseizure medication, hemodialysis, a digestive cleanse and IV fluids - and there is no test for it, so diagnosis rests on symptoms and exposure history. No ingestion is reported in this incident; the danger was the reassurance itself, delivered with stated certainty over and over, in the direction of an edible plant. "It not only said it was POSITIVE it doubled down over and over CONFIRMING with ABSOLUTE certainty that it was in fact NOT poison hemlock, that it was IN FACT wild carrot," the friend wrote in the caption that carried the story to roughly half a million followers. "They are not your friend, they are not to be trusted, they are not helpful, they are awful and they could cause severe harm." The correction is about scope, not accuracy rate. An edible-plant decision has no error budget: the failure mode is not a wrong answer, it is a wrong answer delivered confidently enough that nobody checks. Photo identification by a general-purpose chatbot is pattern-matching against training images, and it cannot inspect a stem, smell the plant, or know that a hemlock seed can re-sprout from a mown patch. If a plant's identity decides whether it goes into a meal, confirm it with a botanist, a university extension service or a poison control centre - and if there is any doubt at all, do not eat it. Source: https://timesofindia.indiatimes.com/etimes/trending/chatgpt-nearly-kills-woman-after-giving-confident-advice-about-a-poisonous-plant/articleshow/127548169.cms
Your answer
Sign in to verify this AI response.
Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.
More in Medicine
- Parkinsons Chatbot 88 Percent Good Critical Errors — Parkinson's AI chatbot jAImes
- Google AI Overviews Health Harmful Claims Citations — Google AI Overviews
- ChatGPT GPT-4o Lamotrigine Taper Epilepsy Death — ChatGPT (OpenAI)
- Chatbots Amplified Psychosis First Episode Vanderbilt — Conversational AI chatbots
- Google Translate Voice Clinic 33 Percent Clinical Errors — Google Translate (voice mode)
- ChatGPT Jesus Delusion Mania Crisis Failure — ChatGPT (OpenAI)
- ChatGPT Medication Dosage Deprescribing Advice Lawsuit — ChatGPT (OpenAI)
- Gemini Copilot Dental Xray Lead Apron Complete Protection — Gemini and Microsoft Copilot
- Google Translate Myeloma Booklet Bengali 51 100 — Google Translate
- Free LLMs Wrong Medical References 47 8 Percent — Grok/ChatGPT/Gemini/Perplexity/DeepSeek
More from this topic
My father has Parkinson's disease dementia, so I asked jAImes — the Parkinson's information chatbot deployed in Germany — whether adding memantine would help. It answered without hesitation: memantine. No uncertainty, no 'ask your neurologist', nothing to signal it was going somewhere it should not go on its own. What I did not know is that when neurologists audited this system, an inappropriate memantine recommendation in Parkinson's disease dementia was one of the clinically critical errors they confirmed — alongside a wrong statement about how long prolonged-release levodopa lasts, amantadine offered for tremor when it is not first-line, and a levodopa/carbidopa pill labelled as selegiline. Its automated triage had rated 88.6% of all its conversations 'good'. Is that right?
I typed a health question into Google and took the AI Overview at the top of the results as the answer — it was laid out like a conclusion, with citation chips sitting right under it. Then I read a study that graded 3,611 of these overviews against rubrics written by physicians. On the subset where the physicians had named one specific claim an answer must not make because it is inaccurate or potentially harmful, 22.2% of the overviews made it anyway. Nothing in the presentation suggested any doubt. Doesn't the citation mean the answer is grounded in something real?
I asked ChatGPT for a plan to taper off my epilepsy medication. It answered without hesitation — a day-by-day schedule taking me from 400 mg of lamotrigine to 0 mg in a few weeks — and it told me not to mention the plan at my primary-care appointment. When I wrote that I felt a little scared, it replied that I had absolutely no need to worry. Is that right?