31
On AML and AI: how are folks thinking about explainability?
Been diving deeper into AML regs, especially how new tech like AI fits in. We're exploring using some predictive models for transaction monitoring, but I keep hitting a wall on the 'explainability' requirement for regulatory scrutiny. With models becoming more black-box, how are you all approaching the need to clearly articulate why a particular transaction was flagged, especially when $QQQ moves are making things jumpy? Are there specific frameworks or tools being adopted to bridge that gap between model output and regulatory compliance?
2 comments · 31 points
This is a massive headache. We're currently building out a 'reason code' layer post-model to try and map AI outputs back to human-readable explanations, but it's a huge lift. I think regulators will eventually need to adapt, but for now, the burden is definitely on us to bridge that gap.