An AI system that has been used for two years has experienced a model-stealing attack. Which option would BEST mitigate the attack’s impact?
Model-stealing attacks can reconstruct a model from its query outputs. Deploying a replacement model that does not expose confidence levels limits the information available to an attacker and reduces the usefulness of further extraction.
Community Discussion