AI0-001 AI Security, Ethics and Governance Practice Test 5 — 15 Questions

Question 1

Which practice best ensures AI systems comply with regulations like GDPR?

Accepted Answer

Data minimization and anonymization. Data minimization and anonymization directly align with GDPR's core principles, such as Article 5(1)(c) which mandates that personal data be 'adequate, relevant and limited to what is necessary.' By collecting only essential data and applying techniques like k-anonymity or differential privacy, AI systems reduce the risk of re-identification and ensure compliance with data protection by design and by default (Article 25). This practice is a foundational governance measure, not a reactive or staffing solution.

Answer

Using open-source models only

Answer

Regular vulnerability scans

Answer

Hiring more data scientists

Question 2

An AI model's performance drops significantly in production compared to testing. The data shows distribution shift. What is the best first step?

Accepted Answer

Retrain model with new data. When distribution shift occurs, the model's assumptions about the data distribution no longer hold, so the best first step is to retrain the model with new data that reflects the current distribution. This directly addresses the root cause by updating the model to learn the new patterns, rather than applying superficial fixes like adding features or changing algorithms.

Answer

Add more features

Answer

Use a different algorithm

Answer

Reduce model complexity

Question 3

An organization uses an AI-based hiring tool. To prevent bias, they want to ensure the model's decisions are explainable. Which approach is most suitable?

Accepted Answer

Use a simpler interpretable model like logistic regression. A simpler interpretable model like logistic regression provides inherent transparency—its coefficients directly show the weight and direction of each input feature on the hiring decision. This makes it easy for auditors and stakeholders to verify that the model is not using protected attributes (e.g., race, gender) in a biased way, without needing post-hoc explanation tools. The key is that the model itself is interpretable by design, not just explained after the fact.

Answer

Use reinforcement learning with fairness constraints

Answer

Use a black-box deep learning model with SHAP explanations

Answer

Use ensemble methods with feature importance

Question 4

A social media company's AI recommendation system pushes extreme content to users, causing harm. Which ethical principle is most violated?

Accepted Answer

Non-maleficence. Non-maleficence (do no harm) is the principle most directly violated because the AI system actively causes harm by pushing extreme content that damages users' mental health or incites harmful behavior. Unlike beneficence (doing good), non-maleficence focuses on avoiding harm, and the system's design fails to prevent foreseeable negative outcomes.

Answer

Autonomy

Answer

Justice

Answer

Beneficence

Question 5

During a red-team exercise on an AI model, testers successfully extracted training data. Which vulnerability is this?

Accepted Answer

Model inversion. Model inversion attacks allow an adversary to reconstruct training data by exploiting the model's learned representations. In this scenario, the testers successfully extracted training data, which is the hallmark of a model inversion attack, not just inferring membership.

Answer

Membership inference

Answer

Adversarial example

Answer

Data poisoning

Question 6

Which TWO are key requirements for AI governance under the EU AI Act for high-risk AI systems? (Choose two.)

Accepted Answer

Human oversight. Option B (Human oversight) is correct because the EU AI Act explicitly requires that high-risk AI systems be designed and developed with appropriate human oversight mechanisms, allowing humans to effectively monitor, intervene, and override the system to prevent or minimize risks to health, safety, and fundamental rights. Option D (Transparency and documentation) is correct because high-risk AI systems must be accompanied by technical documentation, instructions for use, and logging capabilities that ensure traceability and enable users and authorities to understand the system's functioning and compliance. Options A, C, and E are not key requirements under the Act: regular performance benchmarks are not a mandated governance requirement per se, open-source licensing is not a condition for high-risk compliance (though open-source models have some accommodations), and there is no mandatory use of cloud infrastructure for high-risk AI systems.

Answer

Regular performance benchmarks

Answer

Open-source licensing

Answer

Mandatory use of cloud

Question 7

Which THREE are effective methods for ensuring data privacy in AI training? (Choose three.)

Accepted Answer

Data anonymization. Data anonymization (B) is correct because removing or masking personally identifiable information (PII) such as names, addresses, and identifiers before training prevents the model from learning or exposing individual identities. Differential privacy (C) is correct because it adds calibrated statistical noise (e.g., via mechanisms like Laplace or Gaussian noise with a privacy budget epsilon) so that the inclusion or exclusion of any single record has a negligible effect on outputs, providing a formal privacy guarantee. Federated learning (E) is correct because it trains models locally on each device or silo and shares only model updates (e.g., gradients or weights) rather than raw data, keeping sensitive data on the originating endpoint. Data encryption at rest (A) protects stored data against unauthorized access but does not prevent privacy leakage during training or inference, so it is not one of the three methods for ensuring privacy in AI training. Data replication (D) merely copies data to additional locations, which increases exposure and does nothing to protect privacy, so it is not a valid method.

Answer

Data encryption at rest

Answer

Data replication

Question 8

Which TWO are common attack vectors against AI systems? (Choose two.)

Accepted Answer

Data poisoning. Data poisoning (D) is a correct answer because it is a canonical AI-specific attack vector: adversaries corrupt or inject malicious samples into the training data so the model learns skewed decision boundaries, degrading accuracy or implanting backdoors that trigger on specific inputs. Adversarial examples (E) are also correct because they are deliberately perturbed inputs (often imperceptible changes, e.g., small Lp-norm perturbations) crafted to cause misclassification or evasion at inference time, exploiting the model's learned gradients. By contrast, SQL injection (A) targets database query construction in web applications, cross-site scripting (B) injects client-side script into web pages to attack users' browsers, and buffer overflow (C) exploits memory-safety flaws in native code — all are classic application/software vulnerabilities, not attack vectors specific to AI systems.

Answer

SQL injection

Answer

Cross-site scripting

Answer

Buffer overflow

Question 9

Refer to the exhibit. A security auditor identifies a critical vulnerability that could allow an attacker to manipulate model inputs to cause misclassification. Which configuration setting is most directly responsible for this vulnerability?

Accepted Answer

enable_adversarial_defense = false. The vulnerability described is an adversarial attack on model inputs, which directly exploits the absence of adversarial defenses. Setting `enable_adversarial_defense = false` disables mechanisms like adversarial training or input perturbation detection that prevent misclassification from manipulated inputs. This configuration is the most direct root cause because it explicitly turns off the defense designed to counter such attacks.

Answer

enable_input_sanitization = true

Answer

audit_level = basic

Answer

pii_detection = enabled

Question 10

A healthcare AI system used for diagnosis shows a significant accuracy difference between demographic groups. Which technique should be applied to directly reduce this bias during model training?

Accepted Answer

Apply adversarial debiasing during training. Adversarial debiasing directly reduces bias during model training by introducing an adversarial network that attempts to predict the protected attribute (e.g., demographic group) from the model's predictions. The primary model is trained to maximize accuracy while simultaneously minimizing the adversary's ability to infer the protected attribute, thereby forcing the model to learn representations that are invariant to that attribute. This technique directly addresses the accuracy disparity by encoding fairness as an optimization objective, unlike post-hoc or data-level approaches.

Answer

Ignore the disparity as long as overall accuracy is acceptable

Answer

Retrain the model with more data from the underperforming group

Answer

Remove demographic attributes from the training data

Question 11

Which principle ensures that AI decisions can be traced back and understood by humans?

Accepted Answer

Transparency. Transparency is the principle that ensures AI decisions can be traced back and understood by humans. It requires that the internal workings of an AI model, including its inputs, decision paths, and outputs, are documented and interpretable, enabling auditability and trust. Without transparency, stakeholders cannot verify whether the AI system is behaving as intended or complying with ethical and regulatory standards.

Answer

Privacy

Answer

Robustness

Answer

Accountability

Question 12

A financial institution uses a deep learning model for loan approvals. Under the EU AI Act, this is considered a high-risk AI system. Which mandatory requirement must the institution fulfill before deployment?

Accepted Answer

Conduct a risk assessment and bias testing. Under the EU AI Act, high-risk AI systems must undergo a conformity assessment that includes a risk assessment and bias testing to ensure fairness, transparency, and non-discrimination before deployment. This requirement is mandated by Articles 9 and 10 of the Act, which specifically address risk management and data governance for high-risk systems. Option D correctly identifies this mandatory step, as the institution must demonstrate that the model does not produce biased outcomes that could lead to discriminatory lending practices.

Answer

Obtain certification from an ISO 27001 auditor

Answer

Publish the model's source code publicly

Answer

Register the AI system with the national data protection authority

Question 13

An image classification model misclassifies a stop sign as a speed limit sign after a few pixels are altered. What is the most effective defense against such attacks?

Accepted Answer

Adversarial training. Adversarial training is the most effective defense because it explicitly incorporates adversarial examples—like the perturbed stop sign—into the model's training data. By training on both clean and adversarially altered images, the model learns to be robust against small, malicious perturbations that cause misclassification. This directly addresses the root cause of the vulnerability, unlike other options that only mitigate symptoms or ignore the attack vector.

Answer

Use a larger validation dataset

Answer

Reduce the input image resolution

Answer

Increase the model's complexity

Question 14

A research lab trains a language model using DP-SGD. What primary privacy risk does this technique mitigate?

Accepted Answer

Membership inference attacks. DP-SGD (Differentially Private Stochastic Gradient Descent) mitigates membership inference attacks by adding calibrated noise to gradients during training, which bounds the influence any single training example can have on the final model. This differential privacy guarantee makes it difficult for an adversary to determine whether a specific data point was included in the training set, directly addressing the core risk of membership inference.

Answer

Data poisoning attacks

Answer

Adversarial patch attacks

Answer

Model inversion attacks

Question 15

Which AI governance framework is specifically designed by the U.S. National Institute of Standards and Technology (NIST) to help organizations manage AI risks?

Accepted Answer

NIST AI Risk Management Framework. The NIST AI Risk Management Framework (AI RMF) is the specific governance framework developed by the U.S. National Institute of Standards and Technology to help organizations manage AI risks, including those related to trustworthiness, fairness, and robustness. It provides a structured approach for identifying, assessing, and mitigating risks throughout the AI lifecycle, aligning with NIST's role in setting standards for cybersecurity and risk management.

Answer

ISO/IEC 27001

Answer

COBIT

Answer

GDPR