Estimating near-verbatim extraction risk in language models with decoding-constrained beam search
Abstract
Recent work shows that standard greedy-decoding extraction methods for quantifying memorization in LLMs miss how extraction risk varies across sequences. Probabilistic extraction -- computing the probability of generating a target suffix given a prefix under a decoding scheme -- addresses this, but is tractable only for verbatim memorization, missing near-verbatim instances that pose similar privacy and copyright risks. Quantifying near-verbatim extraction risk is expensive: the set of near-verbatim suffixes is combinatorially large, and reliable Monte Carlo (MC) estimation can require ~100,000 samples per sequence. To mitigate this cost, we introduce decoding-constrained beam search, which yields deterministic lower bounds on near-verbatim extraction risk at a cost comparable to ~20 MC samples per sequence. Across experiments, our approach surfaces information invisible to verbatim methods: many more extractable sequences, substantially larger per-sequence extraction mass, and patterns in how near-verbatim extraction risk manifests across model sizes and types of text.
Cite
@article{arxiv.2603.24917,
title = {Estimating near-verbatim extraction risk in language models with decoding-constrained beam search},
author = {A. Feder Cooper and Mark A. Lemley and Christopher De Sa and Lea Duesterwald and Allison Casasola and Jamie Hayes and Katherine Lee and Daniel E. Ho and Percy Liang},
journal= {arXiv preprint arXiv:2603.24917},
year = {2026}
}