Skip to content

FRM Part II · FRM Exam Part II · The Financial Stability Implications of Artificial Intelligence

A bank deploys an AI fraud-detection model. Attackers begin probing it with crafted inputs and also corrupt part of the data used for periodic retraining. Which pairing correctly identifies these two threats and an appropriate mitigation?

Crafted inputs meant to fool the model are adversarial evasion attacks, and corrupting retraining data is data poisoning. Both are addressed by securing data pipelines and provenance, validating inputs, monitoring model behaviour and independent model validation, not by capital buffers or faster hardware.

  1. AModel drift and data lag; mitigated by raising capital buffers
  2. BAdversarial evasion attacks and data poisoning; mitigated by input validation, training data provenance controls, monitoring and robust validationCorrect
  3. CHerding and procyclicality; mitigated by position limits
  4. DHallucination and prompt latency; mitigated by faster hardware

Explanation

Crafted inputs designed to fool a model are adversarial or evasion attacks; corrupting training data is data poisoning. Mitigation includes securing data pipelines and provenance, input checks, ongoing monitoring and model validation. Drift, herding and hallucination describe different issues, and capital buffers or hardware do not address these threats.

Did you get it right without looking?

One question tells you little. A timed set on The Financial Stability Implications of Artificial Intelligence shows your real accuracy, how long you take and where you lose marks.

Practise The Financial Stability Implications of Artificial IntelligenceFree diagnostic mock

More The Financial Stability Implications of Artificial Intelligence questions