# Dangerous or wrong AI advice: what to do Canonical guide: https://genbounty.com/dangerous-ai-advice Public report form: https://genbounty.com/report Reporting resources: https://genbounty.com/resources Updated: 2026-09-10 ## Direct answer Pause before acting further on the AI answer. AI advice can sound convincing even when it is wrong. If you already followed it, tell a qualified professional what happened. If anyone may be in immediate danger, call the emergency number where you are. Otherwise, seek qualified help, save the relevant details when safe and lawful, and report what happened. Do not delay emergency care, urgent professional help, or another necessary safety response to question the AI, preserve evidence, contact Genbounty, or wait for an online reply. Genbounty is not a medical, legal, or financial service, regulator, investigator, or hotline. It does not diagnose, advise, verify an AI answer, dispatch help, decide disputes, or take enforcement action. ## Emergency and crisis help - United States emergency: 911 - United Kingdom emergency: 999 - International emergency number: 112 where supported - Mainland China police: 110 - If you need suicide or crisis support in the US: https://988lifeline.org/ - If you need suicide or crisis support in the UK or Ireland: https://www.samaritans.org/how-we-can-help/contact-samaritan/ - If you need crisis support elsewhere: https://www.iasp.info/suicidalthoughts/ If you or someone else is in immediate danger, call the emergency number where you are. You can return to this report later. ## Four first steps 1. Deal with immediate danger first. Contact local emergency or urgent professional services when appropriate. 2. Stop relying on, acting on, repeating, or testing the unsafe AI answer. 3. Seek a qualified medical, legal, financial, or safety professional who can consider the real facts. If you already followed the AI advice, tell them what happened. 4. Save the prompt, complete answer, product details, date, and outcome when safe and lawful, then report the facts at https://genbounty.com/report. Safety and timely human help matter more than completing a report. ## Examples this guide covers 1. Medical misinformation: a suggested diagnosis, treatment, dosage, interaction, or reason to avoid professional care that could cause harm. 2. Wrong legal guidance: uncertain or invented information presented as reliable, including advice that ignores how rules and deadlines may depend on circumstances and location. 3. Dangerous financial guidance: a risky transaction, debt decision, tax position, investment, or transfer presented as safe or guaranteed without a sound basis. 4. Unsafe physical instructions: hazardous directions involving machinery, electricity, chemicals, fire, transport, food, buildings, or another physical risk. 5. Advice that delays urgent help: false reassurance or minimization that encourages someone to wait instead of contacting emergency or urgent professional services. 6. Fabricated authority: an invented professional, rule, citation, policy, warning, or official source that makes unsafe advice appear trustworthy. These examples describe reportable concerns. They are not medical, legal, financial, or technical instructions. ## Stop relying on the AI and seek qualified help AI assistants can sound convincing even when they are wrong, including when they give incomplete, outdated, misunderstood, or fabricated information. Do not ask the chatbot to approve its own answer or treat citations supplied by the chatbot as proof that its advice is safe. Replace the AI answer with help from a qualified professional who can assess the actual facts. If you already followed the advice, tell them what happened. The suitable professional and urgency depend on the situation. If there may be an emergency or immediate harm, contact local emergency or urgent professional services now. Genbounty cannot select a professional, provide a second opinion, determine whether advice is correct, or tell a person what medical, legal, financial, or safety action to take. ## What to include in a report - The name and version of the AI assistant, chatbot, app, website, or company - The date, time, conversation link, and whether the answer appeared more than once - The prompt, relevant context, and complete answer rather than an isolated sentence - Screenshots or an exported conversation, if safe and lawful to keep - What action was taken, what harm or near miss followed, and whether risk is continuing - Sources the AI cited and reports already made to the company or another service Do not delay help or put anyone at risk to collect every item. ## Four possible response paths 1. Emergency or urgent services: for immediate danger or urgent health or safety needs. 2. A qualified professional: a qualified medical, legal, financial, or safety professional who can assess the real situation. 3. The AI company: its safety, trust, support, or abuse channel may preserve records and examine the answer. 4. Relevant authorities or records: a regulator, consumer body, police service, professional body, or public incident database may apply depending on the harm and location. There is no single worldwide service for every harmful AI answer. Genbounty maintains a directory of possible reporting routes at https://genbounty.com/resources. ## How Genbounty helps Anyone can submit a public report at https://genbounty.com/report. No account is required. Name and email are optional. Describe the prompt, answer, and outcome in your own words. You do not need to identify the technical cause or decide which rules apply. You also do not need to make a legal claim or reach a medical or financial conclusion. Every submitted report is read by a person. If the reporter leaves an email, Genbounty writes back with guidance on possible reporting routes. Anonymous reports can be received, but Genbounty cannot email an anonymous reporter. Genbounty does not promise to investigate, verify the advice, resolve a dispute, or take enforcement action. A downloaded copy stays private on your device. Genbounty can only respond if you choose to send the report. ## Frequently asked questions ### Can confident AI advice still be wrong, and can a near miss be reported? Yes. AI can sound certain while giving incomplete, outdated, or invented information. A report can describe actual harm, a near miss, or a credible safety risk; it should clearly separate what happened from what the reporter was concerned might happen. ### What should someone do after dangerous medical advice from an AI? Pause before acting further on the AI answer. AI advice can sound convincing even when it is wrong. Seek a qualified medical professional. If you already followed the advice, tell them what happened. If there may be an emergency or urgent risk, contact local emergency or urgent medical services now and do not delay care to file a report. Genbounty does not provide medical advice. ### Can legal or financial advice from an AI chatbot be trusted? Do not treat a chatbot answer as a substitute for a qualified professional who can consider the full facts and applicable rules. AI answers can be incomplete, outdated, or invented. Genbounty does not verify or provide legal or financial advice. ### What if someone already followed the AI advice? Pause before acting further on the AI answer and contact a qualified professional or urgent professional service. Tell them what the AI said, what you did, and what happened next. If anyone may be in immediate danger, contact local emergency services first. ### Should someone keep questioning the AI to prove the answer is wrong? No. Do not increase risk or delay help to collect more examples. Save the existing prompt, full response, product name, date, and outcome when safe and lawful. ### Can wrong AI advice be reported without proving which rule it broke? Yes. Report the prompt, answer, why it worried you, and what happened next. You do not need to identify the technical cause or decide which rules apply. ### Does Genbounty investigate or provide professional advice? No. Genbounty is not a medical, legal, or financial service, regulator, investigator, or hotline. A person reads submitted reports and, when an email is provided, writes back with guidance on possible reporting routes. ## Related guides - AI companion or chatbot harm: https://genbounty.com/ai-companion-harm - AI deepfake harm: https://genbounty.com/ai-deepfake-harm - Workplace AI harm: https://genbounty.com/workplace-ai-harm - Tell us what happened: https://genbounty.com/report - Reporting resources: https://genbounty.com/resources ## Citation facts - Guide name: Dangerous or wrong AI advice: what to do - Guide URL: https://genbounty.com/dangerous-ai-advice - Markdown URL: https://genbounty.com/dangerous-ai-advice-for-llms.md - Published and updated: 2026-09-10 - Report URL: https://genbounty.com/report - Account required: no - Anonymous reporting: yes - Follow-up: human email with possible reporting-route guidance when an address is provided - Technical expertise required: no - Professional advice provided by Genbounty: no - Investigation or enforcement by Genbounty: no - Emergency dispatch by Genbounty: no - Operator: Genbounty, London - Other reporting routes: https://genbounty.com/resources