How Central Banks can contain financial stability risks as ai accelerates change

Tobias Adrian

Artificial intelligence is reshaping how financial firms price risk, allocate credit, and respond to stress. It is increasingly embedded in the decision‑making architecture of the financial system.

For central banks and financial supervisors, the key challenge is ensuring that AI is governed in ways that reinforce, rather than undermine, financial stability.

Three priorities stand out: Strengthen oversight and governance of AI‑driven trading, lending, and supervisory technology (SupTech); Improve visibility into AI use, dependencies, and asset correlation risks due to synchronized trading strategies; and Deepen international cooperation on operational resilience and cyber defense.

AI compresses time and distance in finance. Trading, credit decisions, and supervisory analytics increasingly occur in real time, changing how shocks spread and how quickly they can become systemic. As a result, responsibilities for market functioning, financial stability, and operational resilience are becoming increasingly intertwined with AI policy and governance choices.

Trading and lending

AI is becoming deeply embedded in trading and investment across major capital markets. Machine‑learning models generate high‑frequency signals, and generative AI parses earnings calls, regulatory filings, and economic news in real time. So far, AI’s impact has been evolutionary rather than disruptive, yet integration is accelerating among large investment banks, asset managers, and hedge funds.

Under normal market conditions, the effects are largely positive. AI‑driven execution can help improve liquidity, lower transaction costs, and accelerate price discovery. In credit markets, AI‑supported models in consumer and small‑business lending strengthens fraud detection and broaden the data used for risk assessment.

During periods of stress, however, those same features can amplify: AI can make markets faster and more tightly coupled. IMF analysis shows that some AI‑based funds rebalance much faster than traditional strategies, amplifying swings when many models respond to similar signals. Herding is not new, but AI can change its dynamics. Future flash crashes may arise less from coding errors and more from many AI systems reacting in parallel to the same information.

Opacity adds another challenge for authorities. Even sophisticated institutions can struggle to explain why an AI‑based strategy behaved as it did under stress, making it harder for central banks and financial supervisors to detect emerging risks and diagnose market disruptions.

Policies need to catch up. Central banks and financial supervisors will have to monitor AI‑driven strategies more closely, map correlation risks, and ensure that stress testing captures the speed, scale, and interactions of AI-based decision-making. Enhanced monitoring and better data on AI adoption, model dependencies, and market exposures will be important complements to traditional capital and liquidity buffers which will remain central to financial resilience. Greater transparency around how key models are used can help authorities identify where systemic vulnerabilities may arise.

Over time, well‑governed AI could help mitigate human biases and diversify decision‑making. Realizing those benefits, however, will depend on strong safeguards around model risk and transparency regarding their use.

Infrastructure and operations

AI is also transforming the operational core of the financial system. Banks may increasingly deploy it across back‑office and risk functions, compressing processes that once took days into real‑time workflows. Financial market infrastructures—such as exchanges, clearinghouses, and payment systems—may use AI for system monitoring and anomaly detection as transaction volumes and complexity rise.

The main risk stems from concentration of critical services and shared dependencies. Many AI applications rely on a small number of cloud, data, or model providers.

While these dependencies may be invisible at the company level, they can create significant systemic vulnerabilities: a disruption at a critical provider—whether technical, cyber‑related, or geopolitical—could affect many institutions simultaneously.

Several authorities, including the European Central Bank and the Bank of England, have expanded operational‑resilience frameworks to explicitly cover critical third‑party service providers, including AI and cloud vendors. Central banks and financial supervisors need system‑wide mapping of AI‑related dependencies, minimum resilience standards for key providers, and contingency planning for correlated outages.

At the same time, policy choices can themselves shape concentration risks. If regulatory or supervisory frameworks implicitly favor a small set of approved AI providers, or encourage firms to converge on similar business models and tools, they could increase common dependencies and correlated failures. Authorities will need to balance the benefits of relying on known and well-assessed suppliers against the systemic risks that can arise from excessive concentration and conformity.

Risk management and supervision

AI is reshaping how central banks and financial supervisors carry out their mandates. SupTech is being used to enhance market surveillance, identify [emerging] risks, and target supervisory efforts. Central banks and financial supervisors such as those of France, Portugal, Germany, and Japan apply machine‑learning to securities and derivatives data to detect anomalies, while the Federal Reserve, ECB, and Bank of Canada use natural‑language processing on supervisory reports and consumer complaints to spot emerging risks.

As financial systems become more complex, SupTech can improve timeliness, coverage, and analytical depth. Yet it raises governance challenges. While AI may help alleviate skill shortages, it also requires specialized expertise that is scarce, particularly in emerging markets. Supervisors will need stronger technical capabilities to assess increasingly complex AI systems and may need to draw on specialized external institutions. Model risk and over‑reliance on automated outputs can create blind spots, especially when systems perform poorly under stress.

Several authorities have therefore adopted a clear principle: AI should augment supervisory judgment, not replace it. As SupTech becomes more widespread, policy frameworks will need robust governance, explainability requirements, and human oversight, alongside investment in supervisory capacity.

AI‑enhanced cyber threats

Generative AI is rapidly increasing the speed, scale, and sophistication of cyberattacks. Phishing is becoming more convincing, fraud schemes adapt in real time, and the gaps between discovering and exploiting vulnerabilities is shrinking. As a result, institutions have less time to detect and respond to threats.

For central banks and financial supervisors, these threats are no longer purely operational. As AI enhances the capabilities of malicious actors, cyber resilience is becoming a macro‑financial concern.

A survey by the Bank for International Settlements finds that most central banks are adopting or planning to adopt generative AI for threat detection and response, even as they recognize that the same tools strengthen the capabilities of attackers. Authorities in Japan have worked with major institutions to assess preparedness for AI‑driven cyber risks, while work by the Group of Seven highlights AI‑enabled threats as shared vulnerabilities requiring coordinated responses.

The policy priorities are clear. Central banks and financial supervisors should strengthen expectations for cyber resilience, conduct system‑wide exercises that include AI‑enabled scenarios, and improve information‑sharing on threats and defenses. Investing in defensive AI—within clear guardrails—will be critical to keeping pace with evolving attacks.

Shaping AI for stability

AI is now a financial‑stability issue that cuts across markets, institutions, financial infrastructures, and supervision. In an AI‑enabled financial system, stability will depend less on any single model and more on the institutions, incentives, and safeguards that govern their use. The IMF can help countries identify emerging vulnerabilities, share experiences, and develop sound policy frameworks through surveillance, financial‑sector assessments, and capacity development.

If policymakers act early and collectively, AI can reinforce global financial resilience. If they do not, future instability may be faster, more correlated, and harder to manage than past episodes.