Thoughts on AI for flagging suspicious transactions?

Been reading a bit about how some institutions are leveraging AI/ML to better identify AML red flags. Curious to hear if anyone here has direct experience with its efficacy, especially with false positives, compared to traditional rule-based systems.