This is not the "right" way to use AI to kill people.

AI lets you do sigint and treat it a lot more like humint. You can e.g. wiretap everybody a suspected terrorist has called in the last year, transcribe all their conversations and pass them through an AI model which flags anything "concerning."

Unlike traditional approaches, AI can distinguish between "bomb" in the context of playing counter strike, discussing a news report and planning an actual terrorist attack.

It can't do anything a human can't do, but it's orders of magnitude cheaper, especially if you can't outsource the human labor due to natsec concerns.