Thoughts on AI for AML transaction monitoring?

Hey everyone, wanted to get some thoughts from the room on the increasing use of AI/ML in AML transaction monitoring. It feels like the industry is pushing hard in this direction, especially with the sheer volume of data we're all dealing with.

From a risk perspective, I can see the benefits in terms of identifying patterns that traditional rule-based systems might miss, and potentially reducing false positives. However, I'm also mindful of the 'black box' issue and ensuring we maintain adequate explainability for regulatory scrutiny. How are folks validating these AI models for effectiveness and compliance? Are there specific types of solutions or approaches that seem to be gaining traction, particularly for cross-border transactions where the complexity really ramps up? Curious to hear practical experiences.