Parameter Manipulation COASP Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard · select all that apply

   A company's AI-powered customer service chatbot uses a 'temperature' parameter that controls response randomness. The parameter is set in a configuration file. A security consultant is assessing the attack surface. Which of the following are valid attack surfaces for manipulating this parameter? Select all that apply.

   • AThe configuration file itself, if an attacker can write to it.
   • BThe API endpoint that the chatbot uses, if it accepts a temperature override.
   • CThe model's training data, if the attacker can poison it.
   • DThe environment variable that the application reads if the configuration file is missing.
   Check answers Discussion Select an answer first
2. 7expert · hard

   A government agency uses an AI system to screen visa applications. The system reads a 'risk_score_threshold' from a database. A recent audit found that the threshold was changed to a very low value, causing most applications to be approved without review. The agency suspects an insider threat. Which mitigation would be most effective to prevent and detect this type of parameter manipulation?

   • AImplement role-based access control (RBAC) on the database and enable audit logging for all changes.
   • BMove the threshold to a configuration file with read-only permissions.
   • CEncrypt the database and store the encryption key on the same server.
   • DUse a database trigger to automatically revert any change to the threshold.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A machine learning engineer at a healthcare startup discovers that a user can append a small, carefully crafted string to the text input of a medical diagnosis chatbot, causing it to return a different diagnosis for the same underlying symptoms. The engineer suspects the model's tokenizer is being exploited. Which type of parameter manipulation is most directly demonstrated?

   • AAdversarial perturbation
   • BSQL injection
   • CConfiguration tampering
   • DHyperparameter poisoning
   Reveal Solution Discussion Select an answer first
5. 9expert · hard · select all that apply

   A security analyst is reviewing a machine learning pipeline that uses a 'max_features' hyperparameter in a training script. The script reads the hyperparameter from a command-line argument. The analyst wants to identify all the ways an attacker could manipulate this parameter. Select all that apply.

   • AInjecting a malicious environment variable that the script uses as a fallback.
   • BChanging the training data to influence the feature selection.
   • CTampering with the script file that parses the argument.
   • DModifying the command-line argument in the process invocation.
   Check answers Discussion Select an answer first
6. 10application · medium

   A cybersecurity firm uses a malware detection model that accepts a 'scan_depth' parameter. An attacker who knows the API sends a request with 'scan_depth' set to 0, causing the model to skip deep inspection and classify malware as benign. What is the most direct security impact of this manipulation?

   • AThe model's confidentiality is breached.
   • BThe model's availability is reduced because it processes fewer files.
   • CThe model's training data is exposed.
   • DThe model's integrity is compromised, leading to false negatives.
   Reveal Solution Discussion Select an answer first