SECAI Core · phase 11 of 15 · noun

Model theft

A security attack where an attacker reverse engineers the organization's machine learning model to gain access to training data and the algorithm

The Explain card

Plain English
Model theft is stealing an organization's machine learning model, by reverse engineering it, extracting it through queries, or copying the weights, to obtain the algorithm and what it learned from the training data.
Example
An attacker sends a large volume of queries to a paid classification API, records the answers, and trains a copy that performs nearly as well, without paying for any of the research.
Why it matters
Models cost millions to build and often are the product. Defenders protect weight files, throttle query volumes, watermark outputs, and watch for extraction-shaped traffic.
Hook
Why build the engine when you can photocopy it one question at a time?

Word knowledge

How the word is built, where it came from, and what it sits beside in memory.

In a sentence

The security team investigated model theft after repeated API queries systematically mapped the classifier's decision boundaries.

Why these words
  • model Italian modello, a small representation identifies the asset involved
  • theft Old English þīefþ, stealing names the unauthorized taking
Where it came from
Origin
English technical compound combining model and theft
Entered the language
2010s
What changed
Model came to denote a computational system, while theft expanded from tangible property to digital and intellectual assets.
How it is spelled
Pattern
It is an open compound written as two separate lowercase words.
Pattern
The final consonant cluster in theft is spelled ft.
Breaks the pattern
Theft is related to thief and thieve but changes their ie spelling to e and adds t.

Spelled like

  • identity theft
  • data breach
  • trade secret
Broken into chunks
  • model
    • modeling
    • modeler
  • theft
    • thief
    • thieve
What it sits beside

Same subject

Same shape

  • identity theft
  • data theft
  • credential theft

AI security attacks

open noun compounds

Where it sits in the deck

Phase 11: Threat Landscape: Attack Vectors and Adversarial Techniques

With defences named, learn what they must defend against — the full catalogue of attack techniques targeting AI systems, their inputs, outputs, training pipelines, and supply chains.

Neighbours in this phase