Bowman Sounds Alarm on AI Cybersecurity Risks in Banking

Federal Reserve Vice Chair for Supervision Michelle Bowman sounded the alarm on AI cybersecurity risks in banking, highlighting both its benefits and dangers. In a recent roundtable discussion, Bowman pointed out that artificial intelligence can be used to detect vulnerabilities in banking infrastructure, but also poses significant threats if not properly supervised.

Bowman singled out Anthropic's Mythos AI model as an example of the dual nature of AI. This frontier model has both potential benefits and risks, depending on how it is used. Bowman noted that the same model can be repurposed by malicious entities to exploit banking systems if left unsupervised.

The issue of AI-related cybersecurity risks has become a top concern for financial institutions, with large banks such as JPMorgan Chase, Citigroup, Bank of America, and Goldman Sachs grappling with how to prepare for the increasing capabilities of these models. Fed Chair Jerome Powell and Treasury Secretary Scott Bessent have been holding meetings with bank CEOs to discuss these concerns.

Bowman emphasized the need for modernized supervisory approaches to address the challenges posed by agentic AI, where models can autonomously initiate transactions or interact with external systems. She also highlighted the disparity between large and small institutions in terms of their ability to develop advanced AI defenses, and proposed that smaller banks receive targeted support from regulatory bodies.