The PRIMARY goal of data poisoning attacks is to:

Question

Options

• Acompromise the confidentiality of output data from the model
• Bcompromise the confidentiality of model input data
• Cmanipulate the behavior of the model during development
• Dundermine the integrity of the AI system's outputs

• A
  2% (1)
• B
  2% (1)
• C
  7% (3)
• D
  88% (37)

Explanation

Data poisoning involves injecting malicious, corrupted, or adversarially crafted data into a model's training dataset. The objective is to cause the trained model to learn incorrect patterns, leading it to produce wrong, biased, or attacker-desired outputs - undermining the integrity of the AI system's results. The attack targets how the model behaves (its predictions and decisions), not the confidentiality of outputs (A) or inputs (B). While poisoning occurs during the development/training phase, the ultimate goal extends beyond development to affect the model's live outputs (ruling out C as the primary goal).

No community discussion yet for this question.