What a $900M Bank Error Teaches Us About AI Agents - John Feminella, Two Sigma
The $900M Citibank-Revlon error offers a control-failure framework for designing safer AI agents
“it gives you a good idea of why those controls were there in the first place”
Two Sigma AI architect John Feminella uses the 2020 Citibank accidental $900M payment to Revlon creditors as a case study for understanding why agent control systems exist and what happens when they fail. The core argument is that regulated industries offer hard-won lessons about intermediary agent failures that AI practitioners should internalize before deploying autonomous systems. The content is cut off before the practical AI takeaways land, limiting the extractable signal.