What's Happening?
The Bank for International Settlements (BIS) is highlighting the potential for Artificial Intelligence (AI) to introduce systemic risks into the financial system, even when individual AI agents are designed to act prudently. Fernando Restoy Lozano, speaking
at a Cambridge University conference on 'Digital regulation in the area of agentic AI,' stated that supervising AI in finance extends beyond merely overseeing banks' use of AI; it also involves ensuring the resilience of banks and the financial system within an AI-shaped economy. The BIS suggests that traditional supervisory tools, such as capital ratios and static risk assessments, may be insufficient to capture emerging risks posed by AI developments. Instead, supervisors will increasingly need to rely on a broader set of instruments for diagnosis and intervention, placing greater emphasis on supervisory judgment. This judgment, however, must be framed within a robust supervisory framework that ensures transparency, consistency across firms, and stability over time. The BIS has previously warned that widespread use of similar AI systems could lead to issues like herding, liquidity hoarding, and fire sales, echoing concerns about how individually rational actions can collectively destabilize a system, similar to lessons learned from the 2007 global financial crisis.
Why It's Important?
The BIS's warning is crucial for the U.S. financial sector and regulatory bodies as it underscores a fundamental shift in how financial stability needs to be approached in the age of AI. The potential for AI systems, even when individually well-designed, to collectively generate systemic instability poses a significant challenge to existing regulatory frameworks. If thousands of AI-driven financial agents react to market signals at machine speed, they could amplify volatility, leading to rapid and widespread financial disruptions. This could impact U.S. banks, investment firms, and the broader economy by creating unforeseen market crashes or liquidity crises. The emphasis on 'supervisory judgment' suggests that regulators will need to develop more dynamic and adaptive oversight mechanisms, moving beyond rigid, rule-based approaches. This shift could necessitate substantial investments in regulatory technology and expertise, potentially increasing compliance costs for financial institutions. Furthermore, the U.S. financial system, being highly interconnected and technologically advanced, is particularly susceptible to these AI-driven systemic risks, making the BIS's insights directly relevant to safeguarding its stability.
What's Next?
In response to these emerging AI-driven risks, financial regulators in the U.S. and globally are likely to explore new approaches to supervision. This could involve developing 'macroprudential' AI governance, which focuses on the interactions between AI agents and the overall system's stability, rather than just the behavior of individual agents. Potential measures might include AI stress tests that simulate the collective behavior of numerous AI agents, treating common dependencies (like foundational AI models) as systemic exposures, and implementing limits on interaction speed or automatic circuit breakers to prevent cascading failures. Regulators will also need to enhance their monitoring capabilities to detect patterns emerging across populations of AI agents, such as convergence on similar strategies or unusual increases in communication. This will require significant collaboration between central banks, financial institutions, and technology providers to establish robust frameworks and share insights. The development of new regulatory tools and policies will be critical to ensure that the benefits of AI in finance can be harnessed without compromising financial stability.
Beyond the Headlines
The BIS's concerns about AI in finance extend beyond immediate market stability, touching upon deeper ethical and legal implications. The concept that 'good agents can make bad systems' challenges the traditional understanding of accountability in financial markets. If systemic instability arises from the complex interactions of numerous individually rational AI agents, pinpointing responsibility for adverse outcomes becomes significantly more difficult. This could lead to a re-evaluation of legal frameworks concerning liability for AI-driven financial decisions. Moreover, the reliance on 'supervisory judgment' in an AI-shaped economy raises questions about the balance between human oversight and algorithmic decision-making. There's a risk that over-reliance on AI could erode human expertise and judgment, making the financial system more opaque and less responsive to nuanced human intervention during crises. The long-term shift could involve a continuous evolution of regulatory philosophy, moving from micro-level compliance to macro-level systemic resilience, with a strong emphasis on understanding and managing the emergent properties of interconnected AI systems.
