Model Inversion involves what?

• A. Attacker extracts sensitive information by analyzing the model's inputs and outputs.
• B. Attacker causes misclassification by altering inputs.
• C. Data leakage through insecure cloud storage.
• D. The model producing outputs that are culturally biased.

Answer: (A)

Model inversion is about extracting sensitive information from a model by probing its responses. When an attacker inputs queries and observes the model’s outputs or confidence scores, patterns in how the model responds can reveal details about the training data. The attacker is effectively reconstructing or inferring data samples or private attributes that the model learned during training. This kind of leakage happens because the model has memorized or encoded information from its training set in a way that isn’t fully abstracted away from the outputs.

This is why the best choice is that the attacker extracts sensitive information by analyzing the model’s inputs and outputs. The other options describe different security concerns—altering inputs to force misclassification describes adversarial attacks aimed at evasion, data leakage from insecure storage concerns data at rest, and biased outputs relate to fairness or bias issues—none of which describe reconstructing or inferring training data from the model’s behavior.