How AI Could Threaten Financial Stability in a Market Shock

The core risk: synchronization

The ECB’s May 2024 Financial Stability Review special feature says generative AI could have a substantial impact on the financial system, with benefits and risks depending on how challenges around data, model development and deployment are handled at both the firm level and the system level ^[4]. The Federal Reserve has framed AI in similar terms, describing a fast-evolving technology whose risks and benefits are becoming more tangible and clear ^[17].

AI can improve risk assessment, capital and liquidity management, operational efficiency, compliance, customer service and data analysis, according to central-bank and Financial Stability Board material ^[3][4][5]. The stability concern appears when many firms use comparable tools, depend on the same outside services or react to the same signals at the same time ^[4][5][19].

What the ECB is watching

Concentrated AI suppliers

The ECB places particular weight on concentration. A 2024 ECB speech identified a risk that much of the value created by AI could be captured by a few companies that dominate the AI ecosystem ^[3]. In finance, the ECB’s May 2024 stability analysis says widespread AI use combined with concentrated suppliers could make operational risk, including cyber risk, systemic ^[4].

That is a common-point-of-failure problem. If many banks, funds or market infrastructure firms depend on the same model provider, cloud platform or data pipeline, an outage, corrupted update, cyber incident or flawed dataset could affect many institutions at once rather than remaining isolated ^[4][5].

Herding in markets

The market version of concentration risk is herding. A financial stability review in the source set warns that extensive AI use without proper safeguards could contribute to cyber concentration risks, herd behavior and higher market correlations ^[6].

In calm markets, similar AI recommendations may look like efficiency. In a sell-off, they can become procyclical: if many systems recommend cutting exposure, raising liquidity buffers or pulling back from market-making at the same time, the result can be less market depth and sharper price moves ^[6][10].

Model, data and deployment failures

The ECB also emphasizes that AI’s impact depends on data quality, model development and deployment choices ^[4]. That makes AI governance a financial-stability issue, not just an IT issue. A model that works in normal conditions may behave differently during a novel shock, and deployment choices determine whether a flawed output remains an internal warning or becomes an automated action across trading, credit, capital or liquidity processes ^[4].

What the Federal Reserve is watching

The Federal Reserve’s concerns overlap with the ECB’s, but Fed material often frames them through supervision, third-party risk and cyber resilience. A Fed speech on AI in the financial system says supervisors need to ensure that risks are managed as AI capabilities evolve ^[17].

Third-party AI dependence

Federal Reserve research finds that the AI technology gap between small and large banks may be widening and that the diversity of nonfinancial companies serving as third-party AI providers may be limited ^[19]. That points to a concentration problem: smaller firms may depend on a narrow vendor ecosystem, while larger firms may have better access to advanced AI capabilities ^[19].

A separate Federal Reserve paper identifies third-party service providers as a hidden cyber fault line in the financial system, with the potential to create systemic risks ^[20]. Combined with AI supplier concentration, that means a technology vendor can become a transmission channel during stress if many financial firms rely on it ^[4][19][20].

Cyberattacks, deepfakes and fraud

Cyber risk is a major Fed channel. In 2025, Michael Barr said AI-enabled deepfakes can replicate a person’s entire identity and have the potential to supercharge identity fraud; he also said cybercriminals are increasingly turning to generative AI ^[30]. Earlier Fed remarks warned that cyber threats can become more disruptive as technology advances and the financial system becomes more interconnected, and that cyber incidents can generate broader systemic effects ^[28].

During market stress, trust and verification matter. AI-enabled fraud, fake communications or identity attacks do not need to move every asset price directly to become destabilizing; they can disrupt authentication, payments, communications or customer confidence while firms are already trying to process fast-moving information ^[28][30].

AI-mediated decision-making

A Federal Reserve staff paper on generative AI and financial stability notes that humans increasingly rely on AI for information gathering and decision-making, either as a co-pilot or through more autonomous systems ^[1]. Once AI outputs are embedded in trading, liquidity management, risk assessment or banking operations, a model error can be transmitted through actions rather than simply appearing in a report ^[1][4].

How AI could amplify a market shock

A plausible stress path is straightforward.

1. A shock hits. Prices fall, volatility rises, alarming information spreads, or a cyber incident disrupts a key provider. Many institutions process the shock using similar AI tools, data sources or outside vendors ^[4][19][20].

2. AI-driven responses converge. Risk systems may recommend cutting exposure, selling similar assets, raising liquidity buffers or reducing market-making. Stability literature warns that extensive AI use without safeguards can encourage herd behavior and higher correlations ^[6].

3. The feedback loop accelerates. Selling and liquidity withdrawal can push prices down further, which then becomes new input for the next round of risk signals. Policy analysis has warned that AI can amplify wrong-way risk and speed up financial crises ^[10].

4. Common infrastructure becomes a transmission channel. The ECB warns that concentrated AI suppliers can make operational and cyber risk systemic, while Fed research identifies third-party service providers as a cyber fault line ^[4][20].

5. Trust can weaken at the worst moment. Deepfakes, AI-assisted fraud or cyberattacks can undermine authentication and confidence when firms, customers and counterparties most need reliable information ^[28][30].

What would reduce the risk

The safeguards follow from the risk channels. Firms and supervisors need to map common AI dependencies, not only individual models, because supplier concentration can turn firm-level technology choices into system-level vulnerabilities ^[4][5][19]. They also need to test AI systems under stressed conditions, especially where data quality, model design and deployment choices determine whether outputs remain advisory or become automated actions ^[4].

Cyber and third-party resilience are central. The Federal Reserve’s cybersecurity report says its supervisory policies and examination procedures address IT risk management, cybersecurity, operational resilience and third-party risk management ^[21]. The ECB’s analysis points to the same system-level logic: a tool that appears manageable inside one institution can still create fragility if many institutions use it in the same way or depend on the same suppliers ^[4].

Bottom line

The ECB and Federal Reserve are not treating AI as a guaranteed crisis trigger. They are warning that financial-stability risk rises when AI adoption is widespread, suppliers are concentrated, models are difficult to validate and many institutions react to the same signals at speed ^[4][17][19].

In a market shock, AI’s strengths can become liabilities. Speed, scale and optimization can help individual firms respond quickly, but they can also produce correlated selling, reduced liquidity, cyber disruption and faster loss of trust across the system ^[6][20][30].