Advanced AI adoption in financial markets could amplify systemic risk

What's this about?

People disagree about whether smart AI tools could make money markets less safe.

The key test asks if faster, better work brings more good than shared risks.

What supporters say

• AI tools can spread a shock very fast, before people can step in.
• Banks and traders may use the same AI tools, data, or plans, causing them to make the same bad move.
• Hard-to-see AI choices may make it tough for firms and watchdogs to spot risks and fix them.

What critics say

• AI can help firms watch for danger, spot odd moves, and manage risk.
• AI trades may help keep enough buyers and sellers in the market during calm times.

How to read this

The number of points on each side does not show who is right; check how strong the proof is.

The bottom line

AI could make market shocks worse, but we are not sure it will do so every time.

The proof shows real risks, while strong proof also shows AI can help spot and manage them.

Advanced AI could make financial markets more vulnerable to system-wide shocks, but the evidence does not show that it will inevitably do so. The central question is whether efficiency gains outweigh the risks of shared models, limited transparency and faster reactions during periods of stress.

The case for

Shared AI systems could make institutions behave alike, even without direct coordination. Banks and traders may rely on the same data, vendors, signals, goals or model designs. If those systems make similar mistakes or recommend crowded positions, a problem at one firm could spread across the market. Financial-stability authorities identify these common dependencies as a possible route from local errors to system-wide stress. 1 (see Figure 1)

Opacity could make that danger harder to control. Firms, supervisors and counterparties may struggle to understand a model’s assumptions, hidden dependencies or response to a changing market. The IOSCO, IMF and other financial authorities have therefore highlighted explainability, model oversight, outsourcing, data quality and concentration as areas of concern. These reports show that vulnerabilities exist, though they do not measure how much damage AI has actually caused across the financial system. 2

Automation could also transmit shocks more quickly. AI systems can respond to news and to one another within very short periods. That could intensify selling, drain liquidity or spread a mistaken signal before people can intervene. Studies of flash crashes by the Securities and Exchange Commission describe rapid interactions between automated orders and sudden liquidity withdrawals. BIS research has likewise found that algorithmic trading can encourage behavior that worsens market stress. 3 (see Figure 2)

The risk could extend beyond individual models. Many institutions may depend on the same cloud providers, data sources or other infrastructure. A cyberattack, technical failure or faulty shared service could therefore affect several firms at once, turning an isolated problem into a broader disruption.

The case against

The strongest argument against the claim is that AI can improve the ability to detect and manage risk. It may help identify fraud, examine complex relationships and measure systemic exposures earlier. Research has applied AI to nonlinear market interactions and to the measurement of systemic and market-wide risk. These findings suggest that AI could help authorities and firms spot dangers sooner, although they do not prove that adoption will reduce overall systemic risk. 4

Automation may also improve markets in ordinary conditions. Reviews by systematic researchers and the BIS generally link algorithmic systems with better liquidity, lower trading costs and more accurate price discovery when markets are functioning normally. One proposed machine-learning approach aims to reduce noisy or unstable trading signals through risk-aware design. 5 (see Figure 3)

These benefits are important, but they address a narrower question. Better fraud detection or normal-market liquidity does not necessarily show how systems will behave during a sudden crisis, when models may respond in similar ways or withdraw from trading together.

The bottom line

The evidence moderately favours the view that advanced AI could amplify systemic risk under certain conditions, but it does not show that AI adoption will generally cause financial crises. The case for amplification is more coherent because regulators and research identify several clear channels: correlated decisions, weak oversight and rapid shock transmission. However, most of that evidence is prospective or based on earlier forms of automation, including algorithmic and high-frequency trading, rather than direct measurements of frontier AI causing system-wide damage.

The overall effect will depend heavily on how the technology is deployed. Model and provider diversity, strong supervisory challenge, testing during changing market conditions, circuit breakers and limits on automated reactions could reduce the danger. The main uncertainty is whether advanced AI will behave differently from existing automation under stress, and whether governance will be strong enough to prevent local errors from becoming common shocks.