Simple Detector for Model Extraction Attacks

Capture Summary

Recent paper on detecting model extraction attacks against hosted LLM APIs. Search result notes that individual extraction queries often resemble benign requests, and prior evaluations often focus on single-query anomaly scoring.

Relevance

• Latest research for model extraction defense.
• Useful for research on sequence-level behavioral detection rather than per-query classification.

Collection Notes

Collected as latest model extraction defense paper; fill complete authors during ingest.