There are many applications of machine learning models for military purposes, some of which you can see with Ukraine drone warfare for instance. Tech like visual recognition predates much of what people call AI now.
However, this is likely referring to LLM usage to predict stuff, which is utterly stupid because it's a language model. It's like those people who think it's sentient because it recites science fiction from its training to pretend to be sentient.
Part of me hopes this is just a grift, since the reality of ChatGPT helping the military pick targets and make strategic decisions is terrifying. Either way we're just so utterly fucked.