AI is not a person, a guaranteed truth engine, an unbiased judge, a private diary or a replacement for accountable expertise.
Do now: Take one bold AI claim you have heard and rewrite it with a condition, limitation and human responsibility.
Capability depends on the task, data, tool, context and quality of review.Bad mental models create bad decisions. Overconfidence leads to unsafe delegation; exaggerated fear blocks useful learning. Clear boundaries let you use the speed without surrendering the steering wheel.
Not a person
A conversational system can use warm language, remember supplied context and imitate empathy. That experience can feel social, but the system does not share your life, relationships or consequences. Treating it as a person can make disclosure feel safer than it is.
Use human language if it helps you think, but keep the operational truth visible: you are interacting with a service that processes data and produces outputs under rules you may not fully see.
Friendly is a design quality. It is not proof of feelings, loyalty or judgement.
Not a truth machine
Generative AI creates likely outputs. It may blend accurate information, outdated information and invented detail into one smooth answer. Search and retrieval can improve grounding, but sources still need to be opened, read and compared.
For claims that affect money, safety, reputation, grades or public trust, demand a source trail. Check dates, authors, original documents and calculations. If the evidence is unavailable, describe the uncertainty instead of polishing it away.
Not automatically neutral or private
Models learn from human-made data and operate inside human-made products. Historical gaps, stereotypes, language imbalance and design choices can appear in outputs. Fairness is not a switch the model quietly handles on your behalf.
Privacy also depends on the service, account settings, organisation policy and data you provide. Remove unnecessary personal information. For confidential work, use only approved systems and follow the data owner’s rules.
If the input would be dangerous on a public noticeboard, pause before pasting it into a tool.
Not a licence or accountable professional
AI may help prepare questions, explain general concepts or organise documents. It does not become your doctor, lawyer, accountant, engineer, teacher or safeguarding officer. Qualified people must interpret evidence within professional rules and the real situation.
The right question is not ‘Can AI produce an answer?’ It is ‘What level of harm follows if this answer is wrong, and who is qualified to decide?’