It's apparently very hard to scale any exchange system to exclude fraud and prevent false positives in a humane manner. AdSense, Paypal, Google Checkout, eGold, and Prosper.com have all failed to figure out how to do it reliably. You can throw fancy machine-learning techniques at the problem and isolate most of the unusual activity, but the problem then is how you filter that between fraudsters and the genuine good guys.
The basic problem is that if you're a low-margin Internet-scale transactional business, the marginal cost of distinguishing fraud from unusual non-fraud is higher than the marginal benefit from letting the transactions through. And if you're Paypal and you get the float on a bunch of locked funds without the need to following banking regulations, it's definitely not in your best interest to do.
The basic problem is that if you're a low-margin Internet-scale transactional business, the marginal cost of distinguishing fraud from unusual non-fraud is higher than the marginal benefit from letting the transactions through. And if you're Paypal and you get the float on a bunch of locked funds without the need to following banking regulations, it's definitely not in your best interest to do.