Honestly, the "fundamental" fix for the adversarial stuff is just a really tight feedback loop. We lean hard into monitoring—combining customer feedback with internal outlier signals—so we can analyze shifts in real-time. Once we identify a pattern, it’s just about speed.

Sometimes an update means a full AI model retrain, other times it’s just a quick logic or product tweak. In this space, if you can’t iterate faster than the fraudsters, you’ve already lost. (which is the problem of 99% of the companies in the space, they are not able to detect fraud, and if they detect it and they want to push a fix, is already too late)