Charta is on Google Play: practise on the go. Get the app
Charta

Show why wider AI adoption matters for financial stability.

FRM® Part II · Current Issues in Financial Markets · Advances in Artificial Intelligence: Implications for Capital Markets Activities · FRM2.CI.1.c

The idea

Wider AI adoption matters for financial stability because it can create shared vulnerabilities across the system, even when each firm's model looks sound. There are four main channels. The first is third-party dependencies and concentration, since a few providers of cloud services, data and foundation models serve many institutions and a failure at one hits all of them. The second is market correlation, from firms trading on similar signals. The third is cyber risk, including AI-assisted attacks and data poisoning, which corrupts training data. The fourth is model risk, governance and data quality, made harder by opacity and by hallucination in generative models. Stability is a system property, so sound individual models can still add up to correlated behaviour. The usual mistake is judging risk only firm by firm. For the exam, match each scenario to its channel, and note that provider concentration is an operational and systemic risk, not a market-risk one.

A practice question

From the question bank, with the reason each wrong option is wrong.

Which is a financial stability concern linked to many institutions relying on the same few AI service providers?

  1. Provider fees would fall to zero, which would damage bank profitability across the whole sectorWhy not: Fee levels are not the stability concern and nothing suggests fees would fall to zero.
  2. An operational failure at one provider could affect many institutions at once - correct
  3. Providers would have to hold Basel capital against the models they supplyWhy not: Capital rules apply to banks and not to technology providers in this way.
  4. Institutions' models would become more diverse, which would raise correlationWhy not: Common providers reduce diversity; the claim reverses the logic.

Concentration in providers of cloud, data or foundation models creates a single point of failure. A disruption can spread across institutions at the same time.

A harder one

A cloud-hosted large language model used by many dealers for communications monitoring suffers a multi-day outage. Which stability channel does this illustrate?

  1. Market correlation from similar trading strategies across institutionsWhy not: The outage is a service failure, not simultaneous trading on common signals.
  2. Model risk from concept drift in the training dataWhy not: Nothing about the model's accuracy changed; the service was unavailable.
  3. Funding liquidity risk from maturity mismatchWhy not: The case does not involve a funding mismatch; it is a technology dependency.
  4. Third-party dependency and concentration, an operational channel - correct

An outage at a common provider affecting many dealers is third-party concentration risk, a channel highlighted for AI because few firms supply the models and infrastructure.

The hardest tier

A supervisor reasons that because every bank's AI model has passed its own validation, system-wide AI risk is contained. Which response best reflects why this reasoning is flawed?

  1. Validation is never effective for AI models, so passing it tells supervisors nothing at allWhy not: Overstates the case; validation helps but does not address aggregate effects.
  2. AI risk arises only when a model is inaccurate, so any inaccurate model should be banned outrightWhy not: Risk can arise even when models are accurate, from herding and common dependencies.
  3. Stability is a system property: sound models can still produce correlated behaviour in aggregate - correct
  4. Banks validate models only for credit risk, so market risk models remain untested everywhereWhy not: Not true generally, and it misses the system-level issue.

Micro-level soundness does not guarantee macro-level safety. Common models, data and providers create correlation and concentration that individual validation does not reveal.

In the app: 6 practice questions on this outcome, spaced review that brings it back before you forget it.

Start free →

More in Advances in Artificial Intelligence: Implications for Capital Markets Activities