What Is Model Extraction?

Model Extraction — an attack that reconstructs a model's parameters or replicates its behaviour by systematically querying it and observing the outputs.

Also called model stealing, this lets an adversary build a close copy of a proprietary model — or probe it for weaknesses — using only its API responses. It threatens intellectual property and can be a stepping stone to other attacks, which is why query monitoring and rate limiting feature in AI security controls.

Source: Machine-learning security research