Central banks on both sides of the Atlantic are sounding alarms about artificial intelligence as a source of systemic financial risk, with new research and survey data painting a picture of a technology that could reshape how market crises emerge and spread.
AI Architecture as a Source of Fragility
A working paper published by the European Central Bank, produced in collaboration with Stanford University, Deutsche Bundesbank, and the University of Naples Federico II, found that different AI systems behave in sharply divergent — and sometimes dangerous — ways during periods of market stress. The study, titled “Ex Machina: Financial Stability in the Age of Artificial Intelligence,” tested Q-learning algorithms and large language models in a simulated mutual fund redemption scenario modeled on the logic of bank runs.
Q-learning systems, a form of reinforcement learning already used in algorithmic trading, proved especially prone to panic. After experiencing even rare losses, the algorithms developed what the researchers called a “hot stove effect,” triggering excessive redemptions even when economic fundamentals remained sound. Large language models avoided that trap but introduced a different problem: they formed divergent beliefs about what other investors would do, weakening coordination and making market outcomes less predictable.
“AI architecture is a first-order determinant of financial stability,” the researchers concluded.
Fed Survey Shows Surging Concern
Separately, the Federal Reserve’s May 2026 Financial Stability Report, released on May 7, found that half of surveyed market participants now cite AI as a potential threat to financial stability. The figure places AI alongside private credit as a top-tier concern, behind only geopolitical risk, cited by 75 percent of respondents, and oil and gas price volatility at 70 percent. The Fed noted that advances in large language models and agentic AI systems “have introduced new challenges in safeguarding system security for financial institutions, infrastructures and third-party service providers”.
The report also flagged AI-driven disruption as a factor weighing on private credit quality, with potential spillover into broader credit markets.
Central Banks Eye Infrastructure Overhaul
ECB Governing Council member José Luis Escrivá added to the chorus last week, calling on central banks to review the resilience of financial infrastructure in light of AI risks. Escrivá advocated for a proactive reassessment of payment processing, trade settlement, and risk management frameworks, framing AI as a systemic risk rather than merely a productivity tool. The ECB has also launched on-site inspections of European banks focused on generative AI applications in IT operations, legal analysis, and customer-facing tools.
The convergence of these warnings — from an academic simulation of AI-driven bank runs to a central bank survey showing rapidly rising practitioner anxiety — underscores a growing consensus that regulating AI in finance will require looking beyond traditional institutional oversight to the design of the machines themselves.
