ModelRefs / Model Stealing — AI Glossary

Model Stealing — AI Glossary

An attack extracting a proprietary model's functionality by querying its API and training a surrogate model on the query-response pairs.

Overview

Model stealing (Tramèr et al. 2016) reconstructs a model's decision boundaries or capabilities by querying the target API at scale and distilling responses into a local model. Concerns: IP theft, circumventing rate limits/safety, and extracting training data signals. Defenses: rate limiting, output perturbation, and watermarking.

Reference details

Topicsafety
Also known asmodel extraction
Last reviewed2026-06-24

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Model Stealing — AI Glossary.

Frequently asked questions

What is Model Stealing?

An attack extracting a proprietary model's functionality by querying its API and training a surrogate model on the query-response pairs.

Is Model Stealing the same as model extraction?

Yes — model extraction are common aliases for Model Stealing.

What concepts are related to Model Stealing?

Closely related concepts include adversarial attack, llm security, knowledge distillation.