A. Exclusion of stakeholder input in assessments.

B. Operational Critical Threat, Asset, and Vulnerability Evaluation focused on organizational risks.

C. Solely technical vulnerabilities in AI models.

D. Short-term tactical responses over strategic planning.

A. The user interface of the AI application

B. The physical hardware running the AI system

C. The marketing materials associated with the AI product

D. The underlying ML model and its training data.

A. Unauthorized access to sensitive information through compromised plugins

B. Reduced model training time

C. Better integration with third-party tools

D. Improved model accuracy

A. Limiting its use to only high-priority vulnerabilities.

B. Automatically patching vulnerabilities without additional configuration

C. Enabling real-time detection of vulnerabilities with actionable insights.

D. Reducing the need for manual vulnerability assessment entirely

A. input sanitation

B. Adversarial testing involves systematically simulating attack vectors, such as input perturbations or evasion techniques, to evaluate an AI model's robustness and identify vulnerabilities before deployment. This proactive method replicates real-world threats, like adversarial examples that fool classifiers or prompt manipulations in LLMs, allowing developers to observe behavioral anomalies, measure resilience, and implement defenses like adversarial training or input validation. Unlike passive methods like input sanitation, which cleans data reactively, adversarial testing is dynamic and comprehensive, covering scenarios from data poisoning to model inversion. In practice, tools like CleverHans or ART libraries facilitate these simulations, providing metrics on attack success rates and model degradation. This is crucial for securing AI models, as it uncovers hidden weaknesses that could lead to exploits, ensuring compliance with security standards. By iterating through attack-defense cycles, it enhances overall data and model integrity, reducing risks in high-stakes environments like autonomous systems or financial AI. Exact extract: "Adversarial testing is a method where simulation of various attack scenarios is applied to analyze the model's behavior, helping to fortify AI against potential threats." (Reference: Cyber Security for AI by SISA Study Guide, Section on AI Model Security Testing, Page 140-143).

C. Model firewall

D. Adversarial testing

E. Prompt injections

A. Limiting analysis to historical data only.

B. Reducing the use of analytics tools to save costs.

C. Predicting future threats through pattern recognition in large datasets.

D. Increasing data silos to protect information.