AAISM Hard Questions — 25 Difficult Questions

Question 1

An organization is using Google Cloud Vertex AI and suspects an adversarial poisoning attack on a custom training dataset. Which tool should be used to verify the integrity and provenance of the training artifacts?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Vertex AI ML Metadata. Vertex AI Model Garden and Metadata are used for lineage tracking, but specifically for provenance and audit, the Vertex AI ML Metadata service allows tracking of dataset versions and training pipeline execution history.

Answer

Vertex AI Pipelines caching

Answer

Cloud Data Loss Prevention (DLP) API

Answer

Cloud Armor WAF

Question 2

Which THREE practices are necessary for managing 'AI Model Version Control' as part of an overall risk management strategy?

Accepted Answer

Using Git-based tracking for model configuration and code.. Version control ensures reproducibility, rollback capability, and auditability in AI risk management.

Answer

Allowing only one developer to modify the model.

Answer

Storing only the production model artifact.

Question 3

How does 'Continuous Monitoring' differ from 'Point-in-Time Auditing'?

Accepted Answer

Continuous monitoring provides real-time visibility into performance and threats.. Continuous monitoring provides real-time detection of drift or threats, whereas audits are snapshots in time.

Answer

Auditing is only for small companies.

Answer

Monitoring is only for external systems.

Answer

Continuous monitoring is cheaper.

Question 4

Which THREE factors increase 'AI Model Risk'?

Accepted Answer

High model complexity (black box).. High complexity, lack of interpretability, and critical business impact are the top drivers of AI model risk.

Answer

Using a low-latency network.

Answer

Using a well-documented framework.

Question 5

Which THREE security controls are effective against 'Data Leakage' in AI?

Accepted Answer

Data Loss Prevention (DLP) tools.. DLP, access control, and output filtering are the standard triad for preventing sensitive data exposure via AI.

Answer

Using a louder cooling system.

Answer

Increasing the model's training duration.

Question 6

An attacker is performing a 'Model Extraction' attack. What is the most likely goal of the attacker?

Accepted Answer

To steal the intellectual property of the model architecture.. Model extraction (or theft) aims to recreate the functionality or proprietary logic of a model by querying it and training a substitute model.

Answer

To gain access to the raw training database.

Answer

To cause the model to crash.

Answer

To inject malicious weights into the model.

Question 7

A researcher is using 'Adversarial Training' to defend against evasion. How exactly does this work?

Accepted Answer

It trains the model on adversarial examples alongside clean data.. Adversarial training includes adversarial examples in the training set, allowing the model to learn to classify them correctly.

Answer

It uses a secondary model to detect malicious input.

Answer

It restricts input to a whitelist of authorized users.

Answer

It adds noise to the weights of the model.

Question 8

You are assessing the risk of 'Data Poisoning'. Which stage of the machine learning lifecycle is most vulnerable to this attack?

Accepted Answer

Training data collection and ingestion.. Data poisoning occurs during the training data ingestion phase, where malicious samples are injected into the training set.

Answer

The model design and architecture selection.

Answer

Model monitoring and logging.

Answer

Inference and deployment.

Question 9

Which THREE documents are vital for an AI audit trail?

Accepted Answer

Model validation and testing reports.. Model version, training set, and validation report are the three pillars of a reconstructible audit trail.

Answer

The office lease agreement.

Answer

A copy of the employee handbook.

Question 10

Which THREE items should be included in an 'AI Inventory'?

Accepted Answer

Unique model identifier and version.. Model version, owner, and risk rating are the core pieces of information needed to manage an AI portfolio.

Answer

The model's favorite color.

Answer

The developer's favorite food.

Question 11

Which THREE components are critical for an 'AI Incident Response' plan?

Accepted Answer

Communication channels with data scientists and legal teams.. An AI IR plan must account for model-specific issues like unexpected performance drops, bias, and adversarial attacks.

Answer

Requirements to publish all internal vulnerabilities publicly.

Answer

Automated deletion of all logs.

Question 12

An organization is performing threat modeling for a RAG (Retrieval-Augmented Generation) pipeline. Which specific vulnerability is unique to the retrieval component?

Accepted Answer

Database poisoning within the retrieved document index.. In RAG, if the search or retrieval index is manipulated, the LLM will generate answers based on attacker-supplied context.

Answer

Overfitting of the LLM on the training set.

Answer

Gradient-based evasion attacks on the LLM.

Answer

Prompt injection targeting the LLM model weights.

Question 13

Which THREE elements are essential for 'AI Governance'?

Accepted Answer

Defined AI ethics and safety policies.. AI governance requires clear policies, oversight committees, and defined risk frameworks.

Answer

Access to all employees' private emails.

Answer

A mandate to use the most expensive GPUs available.

Question 14

What is 'Model Stealing' and why is it considered a security incident?

Accepted Answer

It is the theft of intellectual property via query analysis. Model stealing involves querying a model to extract enough information to replicate its functionality, leading to the loss of proprietary intellectual property.

Answer

It is the slow down of the model due to heavy load

Answer

It is the unauthorized access to the server room

Answer

It is the accidental deletion of model files

Question 15

What is the 'Membership Inference' attack in machine learning?

Accepted Answer

Determining if a specific record was in the training set. This attack determines whether a specific data point was included in the training set, potentially compromising privacy if the data is sensitive.

Answer

Brute forcing the user password

Answer

Adding noise to the output

Answer

Stealing the model's entire weights file

Question 16

Which THREE factors must be included in a 'Model Risk Management' (MRM) policy?

Accepted Answer

Clear model lifecycle stages and governance gates.. A robust MRM policy covers the lifecycle from model development and validation to monitoring and retirement.

Answer

The names of all employees who use the AI.

Answer

A list of all model research papers published by the company.

Question 17

Which THREE technical controls effectively manage 'Third-Party AI Vendor' risks?

Accepted Answer

Independent verification of the vendor's model performance.. Managing third-party AI risk involves technical validation, contractual oversight, and ongoing monitoring.

Answer

Asking the vendor to delete all their internal logs.

Answer

Requiring all employees to use the vendor's internal Wi-Fi.

Question 18

When establishing a risk management framework for GenAI, which THREE factors should be prioritized to satisfy the NIST AI Risk Management Framework requirements?

Accepted Answer

Continuous monitoring of model performance metrics.. NIST AI RMF prioritizes trustworthiness, safety, and reliability through continuous monitoring and governance.

Answer

Requiring all models to be open-source.

Answer

Implementing a strict 'no external API' policy for all models.

Question 19

You are auditing an AI system for 'Model Explainability' (XAI). Why is XAI critical from a risk management perspective?

Accepted Answer

It provides transparency for accountability and bias detection.. Explainability allows organizations to understand the 'why' behind model decisions, essential for regulatory compliance and debugging bias.

Answer

It is required to make the model run faster.

Answer

It encrypts the input data.

Answer

It enables models to train without labels.

Question 20

Which of the following describes the purpose of 'Differential Privacy' in a machine learning system?

Accepted Answer

To prevent the identification of individuals in the training data. Differential privacy ensures that an individual's data cannot be inferred from the aggregate model output by injecting statistical noise during training.

Answer

To enable cross-cloud model training

Answer

To allow for better interpretation of model layers

Answer

To ensure the model is faster at inference time