Can an ML model actually avoid making false positives and how do you know your model isn`t just memorising patterns ?

Hello ,Im a self taught developer in python who took interest in machine learning very curious about how an ml model actually works. I have built a double layered ai agent with an xgbooost model and audit layer of gemini . But it has brought to my attention that some of the predictions made by the model are actually false positives . And really how do you know your model isnt just memorising patterns and even so what if now when you test it with something it can`t understand and it just gets something from your dataset and it makes a false prediction. And know tis model is used in production and leads to losses. Since my models was working with medical data about drug to drug interaction side effects i had to make sure in case it makes a false prediction since now it could false categorise a toxic drug combination as safe. I thought this should be the best practice in machine learning models to be able to flag out the false positives(MOST IMPORTANT) and false negatives.