liveBook · Manning

Model extraction

As explained in AI Governance

An attack that reconstructs or approximates a target machine-learning model by querying it and analyzing its outputs.

Model extraction reconstructs or approximates a private model through repeated API queries. An attacker studies the responses and trains a student model to mimic the target’s behavior; the result need not reproduce its weights exactly to expose its functional logic. section 4.4.3

Unlike direct theft, which accesses stored model weights, extraction targets the model’s behavior through its interface. This distinction matters because a secured storage environment does not by itself prevent functional imitation. section 4.4.3

Reasoning-model imitation

Sky-T1-32B-Preview was reported to perform on par with o1-preview on popular reasoning and coding benchmarks using under $450 in compute. OpenAI and Microsoft also investigated whether DeepSeek used large volumes of OpenAI API data to train competing models. section 4.4.3

How AI Governance covers it

How deep it goes
Explains the query-and-train process and distinguishes functional imitation from weight theft, without detailing extraction defenses.
Examples
Cites Sky-T1-32B-Preview’s reported benchmark performance and investigations into DeepSeek’s use of OpenAI API data.
What you can do after
Understand how repeated API access can reveal functional logic without exposing the original model weights.
How it connects
Connects model extraction to model distillation, API security, and intellectual-property theft.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.