FRM Exam Part II · Advances in Artificial Intelligence: Implications for Capital Markets Activities

Systemic and Financial Stability Risks of AI in Capital Markets

Updated 11 October 2026 · Fact-checked

AI creates systemic risk when many firms use similar models, data or providers. This causes herding, correlated trades, concentrated dependence on a few vendors, faster and sharper volatility, flash crashes, and shared cyber and operational failures. To solve questions, find the shared channel, then link it to the system-wide effect.

Understand Systemic and Financial Stability Risks of AI

Start with a simple idea. One firm's mistake is a firm problem. When many firms make the same mistake at the same time, it becomes a systemic problem. AI raises this risk because it pushes firms toward the same tools, data and suppliers.

Herding and correlated strategies: if many firms train models on similar data, or buy models from the same vendor, their signals look alike. When a shock arrives, they may all sell the same assets together. Selling pushes prices down, which triggers more selling. This is a feedback loop. Diversity of views normally absorbs shocks. Uniform models remove that cushion.

Concentration in third-party providers: most firms do not build the largest models or the cloud infrastructure themselves. They rely on a small number of cloud, data and model vendors. If one provider fails, is hacked or has a flawed model update, many institutions are hit at once. This is a single point of failure. Switching is hard and slow, so the dependence is sticky. A related point is that the vendor's own suppliers (fourth parties) add hidden links.

Market volatility and flash crashes: AI-driven trading acts in milliseconds. Fast, similar algorithms can withdraw liquidity together, widen spreads and cause rapid price falls that may reverse quickly. Models trained on calm periods may behave unpredictably in stress they have not seen. Limited explainability makes it hard for humans to spot or stop the problem in time.

Cyber and operational risks: AI widens the attack surface. Attackers can use AI for more convincing phishing, deepfakes and faster attacks. Models can also be attacked directly, for example through poisoned data. Shared platforms mean one incident can spread. For FRM Part II, the Financial Stability Board view is that these vulnerabilities (third-party dependence, market correlations, cyber risk, and model risk with data quality and governance) are the main channels to monitor.

How to solve Systemic and Financial Stability Risks of AI questions

There is no calculation here. Use the same reasoning chain on every scenario question.

1Read the scenario and identify what is shared: model, data, vendor, cloud, or trading signal.
2Name the channel: herding or correlated strategies, third-party concentration, volatility and flash crash, or cyber and operational.
3Explain the transmission: how one trigger becomes many firms acting or failing together.
4Check whether the risk is firm-level or system-wide. Systemic answers stress simultaneity and feedback loops.
5Look for the amplifier: speed, opacity, low model diversity, high switching costs, or thin liquidity.
6Match the mitigant to the channel: vendor diversification and exit plans for concentration, human oversight and circuit breakers for flash crashes, stress tests and monitoring for correlation.
7Eliminate options that are overstated, such as saying AI always causes crashes, or that confuse the channels.

Quickest way: Shared-thing test

When to use it: Use it for any multiple-choice scenario under time pressure.

Ask: what do the firms have in common?
Same vendor or cloud points to concentration risk.
Same model or data points to herding and correlated trades.
Speed and sudden liquidity withdrawal points to flash crash or volatility.
Malicious actors or compromised models point to cyber risk.
Pick the option that describes simultaneous, system-wide impact.

Common mistakes in Systemic and Financial Stability Risks of AI

Treating herding and third-party concentration as the same risk.
Both involve many firms behaving alike.
Fix: Herding comes from similar models or data driving similar trades. Concentration comes from dependence on the same provider for a service.
Saying AI always increases volatility.
Students overstate the headline risk.
Fix: AI can improve efficiency and liquidity in normal times. The concern is that uniform behaviour can amplify stress. Use words like can and may.
Assuming diversifying across many AI models removes concentration risk.
Students ignore the shared underlying provider or data.
Fix: Different tools built on the same vendor, foundation model or cloud still share a single point of failure.
Treating a flash crash as only a model error inside one firm.
Focus on the firm's own control failure.
Fix: A flash crash is a market-wide event driven by interacting fast algorithms and liquidity withdrawal. Firm controls help, but the system effect is the point.
Ignoring explainability and monitoring when choosing mitigants.
Students pick technical fixes only.
Fix: Opacity makes problems harder to detect. Good answers include governance, human oversight, testing and data-quality controls.
Forgetting cyber risk is both AI-enabled attacks and attacks on AI systems.
Only one side is remembered.
Fix: Cover malicious use of AI by attackers and vulnerabilities in the models, data and platforms firms use.

Worked examples

Example 1

Five large asset managers buy risk-signal models from the same vendor. A sudden shock triggers sell signals at all five on the same day. Which systemic risk channel is most directly illustrated, and why?

Show the solution

Shared thing: the same vendor model produces the same signal.
Channel: this is herding through correlated strategies, since the trades come from similar model output.
Transmission: simultaneous selling pushes prices down, which can trigger further sell signals, a feedback loop.
Also note: the common vendor adds concentration risk, but the stated outcome is simultaneous trading, so herding fits best.

Answer: Herding and correlated strategies. Similar model signals cause simultaneous selling that amplifies price falls through a feedback loop.

Example 2

A bank relies on one cloud provider to host its AI-based credit and trading models. The provider has an outage lasting several hours. Which risk is this, and which mitigants fit best?

Show the solution

Shared thing: a single provider supports critical functions.
Channel: third-party concentration, a single point of failure. If many banks use the same provider, the impact is system-wide.
This is operational in form, but the systemic element comes from the shared dependence.
Mitigants: assess criticality of the provider, use multi-provider or multi-region arrangements where feasible, define exit strategies, test business continuity plans, and map fourth-party dependencies.

Answer: Third-party concentration risk, an operational and systemic vulnerability. Best mitigants are provider diversification where feasible, exit and continuity plans, testing, and mapping of fourth-party dependencies.

Exam tips

Questions are usually scenario-based. Identify the shared model, data or provider first.
Learn the four FSB-style channels: third-party dependence, market correlations, cyber risk, and model risk with data quality and governance.
Watch for absolute words such as always or eliminates. They usually mark wrong options.
For mitigants, match the control to the channel instead of picking a generic answer.
Link to liquidity: flash crashes involve liquidity withdrawal, so connect them to market liquidity ideas.

Practice questions from Advances in Artificial Intelligence: Implications for Capital Markets Activities

Systemic and Financial Stability Risks of AI: frequently asked questions

How can AI cause herding in markets?

If firms use similar models, training data or vendors, their signals and trades become alike. In stress they may buy or sell together. This reduces diversity of views and can amplify price moves.

What is third-party concentration risk in AI?

It is the risk that many institutions depend on the same few providers of models, data or cloud services. A failure, cyberattack or flawed update at one provider can hit many firms at once. Switching providers is usually slow and costly.

Can AI cause a flash crash?

It can contribute. Fast, similar algorithms may withdraw liquidity together, widen spreads and push prices sharply down in minutes. It is a possible amplifier, not a certainty, and normal-time effects may be positive.

How do I answer AI financial stability questions in the FRM exam?

Find what the firms share, name the channel, then explain how a single trigger becomes a system-wide effect. Finish by checking the mitigant matches the channel.