On September 15, 2008, a mathematical model that had passed every regulatory test destroyed the global economy in a single weekend. This is the story of how we built the rules that were supposed to prevent that from happening again, and why those same rules now govern every AI system touching your money.
We begin in the wreckage. Lehman Brothers is dead. Tim Geithner is watching screens at the New York Fed as the numbers turn impossible. The Value-at-Risk models that every bank trusted, that every regulator approved, are spitting out predictions that miss reality by orders of magnitude. The models worked perfectly. The models were catastrophically wrong.
This episode traces the five layers of governance that now wrap around every algorithm in American banking. We start in 1863, when Abraham Lincoln needed to fund a war and Hugh McCulloch built the first national banking system from scratch. We move through Jekyll Island and the birth of the Federal Reserve, through the bank runs of 1933 that created the FDIC, through Ruth Bader Ginsburg's quiet revolution in civil rights law that made algorithmic fairness a federal requirement decades before the term existed.
We meet the regulators who wrote SR 11-7, the obscure 2011 guidance document that became the bible of model risk management. We watch the CFPB demand that banks explain their AI decisions in plain English. And we arrive at November 30, 2022, when ChatGPT launched and broke every assumption the governance stack was built on.
The five layers were designed to prevent the next 2008. But the algorithms they were meant to control have evolved faster than the rules. A journey through 160 years of banking law, mathematical failure, civil rights litigation, and the uncomfortable question of whether we can govern what we cannot fully understand.