Gary Gensler says it's “nearly unavoidable” that AI will cause a financial crisis if many institutions rely on the same underlying base model or data aggregator
Context & Ripple Effects
Gensler’s warning broadens the SEC’s earlier AI agenda beyond adviser conflicts: the agency was already considering rules for AI-driven recommendations in financial advisers’ and brokerages’ use of AI. The issue here is not merely biased advice, but correlated failure when institutions share a technical dependency.
Later coverage from European banking watchdogs similarly characterized frontier models as a systemic risk for lenders, reinforcing the arc from conduct-focused AI oversight toward resilience and concentration risk.
First-order effects
- Financial firms using the same base model or data source face sharper scrutiny of whether a common error, outage, manipulation, or market signal could drive synchronized decisions.
- AI-model and data-aggregation providers become more consequential counterparties to finance, because their operational failures can propagate across multiple institutions at once.
Second-order effects
- Banks, brokerages, and advisers have incentives to map shared model and data dependencies, add fallback processes, and avoid allowing one provider to determine critical decisions across the organization.
- Providers competing for financial-services customers may need to demonstrate reliability, transparency, and portability—not only model performance—as buyers weigh concentration exposure.
Third-order effects
- If common AI dependencies become embedded in financial workflows, model concentration risk could become a prudential issue alongside traditional third-party and operational-risk management.
- The regulatory focus is likely to move from individual AI outputs to system-wide correlation: whether many firms can fail or react together because they rely on the same underlying infrastructure.
The trend: AI oversight in finance is evolving from concerns about individual automated decisions toward the systemic consequences of concentrated model and data infrastructure.