Improving Supervision Through AI

In a Financial Times opinion piece published late last month, Amit Seru, Professor of Finance at Stanford Graduate School of Business, argues that AI could help counter the institutional incentives and cultural biases that can discourage bank supervisors from acting on emerging risks.

Seru argues that supervision requires decisions under uncertainty, but the incentives surrounding those decisions are asymmetric. Acting unnecessarily can expose supervisors to criticism from politicians or bank management and carry significant career costs. Delaying action, on the other hand, may attract little immediate consequence if the risk dissipates or only becomes apparent much later. The result, he argues, is a “bias towards forbearance or inaction.”

He points to the 2023 collapse of Silicon Valley Bank (SVB), where supervisors had identified significant vulnerabilities well before the bank failed but did not take sufficient remedial action. Seru argues that the central problem was not a lack of information but the willingness to act on it. He contends that recent Federal Reserve reforms, including clearer lines of responsibility, fewer supervisory layers, and greater focus on material risks, address some of the institutional sources of delay but do not fully resolve the incentives favoring inaction.

Seru proposes using AI to analyze supervisory information and flag material deterioration in banks’ risk profiles. Supervisors could still decide not to intervene, but would be expected to explain that decision, making inaction more visible and accountable rather than allowing it to remain the default. He stresses that AI should support, rather than replace, supervisory judgment.

Technology alone, however, would not solve the problem. Seru argues that supervisors also need institutional protection when they make reasonable decisions under uncertainty that later prove mistaken. Without greater tolerance for good-faith supervisory judgment, he suggests, efforts to improve supervisory tools and organizational structures may do little to strengthen the willingness to act.