Parameter Manipulation COASP Practice Questions (Page 8)

Questions 36–40

1. 36application · medium

   An e-commerce company uses a recommendation system that takes user profile parameters (e.g., age, location, past purchases) as input. A penetration test reveals that an attacker can manipulate the 'past_purchases' parameter to include items they never bought, causing the model to recommend expensive products. The company wants to prevent this without affecting legitimate users. Which mitigation strategy is most appropriate?

   • AUse a different recommendation algorithm that is less sensitive to input changes.
   • BAdd a CAPTCHA to the recommendation request to ensure a human is making the request.
   • CRate-limit the recommendation API to reduce the number of requests an attacker can make.
   • DImplement server-side validation to ensure that the 'past_purchases' parameter matches the user's actual purchase history from the database.
   Reveal Solution Discussion Select an answer first
2. 37application · medium

   A healthcare startup uses a containerized AI model for medical image classification. The model reads configuration parameters from a YAML file that is mounted into the container at runtime. A developer accidentally exposes the container's management API to the internet. An attacker discovers they can modify the mounted YAML file through the management API, changing the model's confidence threshold to 0.0. What is the most likely immediate impact of this parameter manipulation?

   • AThe model will crash because a threshold of 0.0 is invalid.
   • BThe model will classify every image as the most probable class, regardless of actual content.
   • CThe model will return an error for every image because the threshold is out of range.
   • DThe model will become more accurate because it will consider more candidates.
   Reveal Solution Discussion Select an answer first
4. 38expert · hard

   A company deploys an AI model for real-time credit scoring. The model's decision threshold is stored in a configuration file that is read at startup. The security team wants to prevent tampering, but the operations team needs to be able to adjust the threshold without restarting the service. Which solution best balances security and operational flexibility?

   • AHardcode the threshold in the application code and require a code deployment to change it.
   • BExpose the threshold as an environment variable that can be changed at runtime without restarting, and rely on network security to protect the environment.
   • CUse a secure parameter store with an API that requires authentication and authorization, and have the application fetch the threshold dynamically at each request.
   • DStore the threshold in a signed configuration file and require a restart to apply changes, with a change-management process for updates.
   Reveal Solution Discussion Select an answer first
5. 39foundation · easy

   In the context of AI systems, what is parameter manipulation?

   • AReplacing the entire AI model with a different model architecture
   • BEncrypting the model weights to prevent unauthorized access
   • CDeleting the training dataset to force the model to retrain from scratch
   • DChanging the values of inputs, configuration settings, or model hyperparameters to alter the behavior of the AI system
   Reveal Solution Discussion Select an answer first
6. 40application · medium

   A company uses an AI model to automatically moderate user-generated content. The model's 'strictness' parameter is stored in a configuration file on a shared network drive. An attacker with write access to the drive changes the 'strictness' from 0.9 to 0.1. What is the most likely impact on the moderation system?

   • AThe system will allow more inappropriate content to pass through, increasing the risk of harmful content being published.
   • BThe system will crash because the strictness is too low.
   • CThe system will flag more content as inappropriate, leading to many false positives.
   • DThe system will operate normally because the strictness is still within the valid range.
   Reveal Solution Discussion Select an answer first