Mark Chu-Carroll has an explanation of what went wrong in the financial institution's risk assessment procedures, which I find a bit hard to believe, but only a bit. If true, it means that these highly-paid technical portfolio analysts were making the kind of mistake you drill out of undergraduates in the second week of an introductory probability course, namely, assuming the independence of evevents (mortgage defaults) that are in fact highly correlated. This seems way too simple an explanation, but maybe it really is as stupid as that.
The slightly more complex version of this is that they knew this calculation was bogus, but were under incentives that promoted the collections of short-term gains at the expense of long-term risks, and somehow shoved this knowledge under the rug.
Also caught a bit of this weekends This American Life episode, which talked to some Wall Street types about what was going on. Sounded interesting but I didn't catch the whole thing and it isn't on the web yet.
Nassim Nicholas Taleb is getting a boost in his reputation from all this, but I'm not sure exactly why. This particular meltdown was not a "black swan", in that it was entirely predictable. But perhaps his distinction between Mediocristan and Extremistan is a key to the theory above. Normally, mortgage defaults are uncorrelated, but in extreme cases they suddenly become highly correlated.
More Taleb here.