There's at least two considerations that come to mind. The first is whether using AI is legally in the clear. Machine learning uses large amounts of data, some if which is copyrighted. When generating text the AI can sometimes reproduce text it has been trained on verbatim, which may constitute a copyright violation if that's not allowed by a license. There's also the question of whether the data was allowed to be used for training in the first place. The same issues are currently being fought about for code generation and image generation.
The other big question is whether current AI language models are fit for purpose. That, of course, depends on what purpose you want to put them to. Spell checking and grammar checking have been a big help to many people for years now. Predictive text on phones has made writing text messages faster. If you consider AI as a further advancement of this trend, then it can help you write what you want faster by better predicting words, sentences and paragraphs. It's just a tool for getting your thoughts to paper faster.
What it's not fit for, is writing the content for you. AI doesn't understand what it writes and just generates a statistically probably sequence of words in a given context. But "probable" doesn't mean it's good, or factual, or anything. Without a writer controlling the process and understanding what's being produced you may not get a well-written end product that makes sense and is factually accurate.
One thing to be especially cautious about is that AI can very confidently write assertions that are wrong, and if you don't know the topic well enough to know better those mistakes can end up in your text. Another important problem is that AI has a general tendency to enlarge biases in its training data, so e.g. it may write in a way that reinforces stereotypes. For these and other reasons, you can't give AI free rein to just write long swathes of text as it will. You always have to check, and take responsibility for every word in the end product.
