Right. That's the author's responsibility, whether he or she is using ChatGPT, a human colleague, an old-school tool like Grammarly or classic Google Translate, or even older-school tools like paper dictionaries.
In all those cases, mistakes can be made and unintended meanings can creep in. The author has to check the output, definitely, but that doesn't mean those tools shouldn't be used at all.
Using such a dictionary instead of the control, a human translator or interpreter, is likely to suffer from the same pitfalls: not knowing a language means you are more susceptible to making errors when the language should be nuanced, and scientific & academic papers seem to be to be nuance-critical productions when it comes to wording
In all those cases, mistakes can be made and unintended meanings can creep in. The author has to check the output, definitely, but that doesn't mean those tools shouldn't be used at all.