AI Incident Response and Forensics COASP Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   After a model poisoning incident, a company has removed the poisoned samples and retrained the model. However, the company is concerned that the retrained model may still have residual vulnerabilities. Which recovery step is MOST effective in ensuring the model is secure?

   • AKeep the compromised model in production while monitoring its behavior
   • BIncrease the model's complexity to make it more robust against future attacks
   • CDeploy the retrained model immediately to restore service as quickly as possible
   • DConduct a thorough validation of the retrained model using a holdout dataset that was not part of the training process
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   After a data poisoning incident, a company has removed the poisoned samples and retrained the model. To prevent recurrence, which additional measure should the company implement?

   • AImplement integrity checks on training data and monitor for unexpected changes in model behavior
   • BRestrict access to the model's API to only internal users
   • CSchedule regular model retraining to ensure the model stays up to date
   • DIncrease the model's complexity to make it more resistant to poisoning
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A company's facial recognition model has been attacked using adversarial perturbations, causing it to misidentify individuals. The incident response team has confirmed the attack and is assessing the impact. Which factor is MOST important to evaluate when determining the impact on model integrity?

   • AThe cost of retraining the model with additional adversarial samples.
   • BThe amount of sensitive data the model may have exposed during the attack.
   • CThe duration of time the model was unavailable during the attack.
   • DThe extent to which the model's outputs are incorrect for legitimate inputs, such as misidentifying authorized individuals.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   Which of the following is a preventive measure that can be implemented during the remediation phase to avoid recurrence of a model poisoning attack?

   • AIncreasing the model's inference speed
   • BDeleting all training data
   • CImplementing integrity checks on training data and model artifacts
   • DDisabling all logging
   Reveal Solution Discussion Select an answer first
6. 5foundation · easy

   Which of the following should be included in an AI incident response playbook?

   • AA copy of the organization's financial statements
   • BA list of all employees' personal contact information
   • CThe organization's marketing strategy
   • DSpecific detection indicators and response actions for a given AI threat
   Reveal Solution Discussion Select an answer first