@inproceedings{kaneko-baldwin-2025-investigating,
title = "Investigating How Pre-training Data Leakage Affects Models' Reproduction and Detection Capabilities",
author = "Kaneko, Masahiro and
Baldwin, Timothy",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.emnlp-main.1201/",
doi = "10.18653/v1/2025.emnlp-main.1201",
pages = "23545--23555",
ISBN = "979-8-89176-332-6",
abstract = "Large Language Models (LLMs) are trained on massive web-crawled corpora, often containing personal information, copyrighted text, and benchmark datasets. This inadvertent inclusion in the training dataset, known as data leakage, poses significant risks and could compromise the safety of LLM outputs. Despite its criticality, existing studies do not examine how leaked instances in the pre-training data influence LLMs' output and detection capabilities. In this paper, we conduct an experimental survey to elucidate the relationship between data leakage in training datasets and its effects on the generation and detection by LLMs. Our experiments reveal that LLMs often generate outputs containing leaked information, even when there is little such data in the training dataset. Moreover, the fewer the leaked instances, the more difficult it becomes to detect such leakage. Finally, we demonstrate that enhancing leakage detection through few-shot learning can help mitigate the impact of the leakage rate in the training data on detection performance."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kaneko-baldwin-2025-investigating">
<titleInfo>
<title>Investigating How Pre-training Data Leakage Affects Models’ Reproduction and Detection Capabilities</title>
</titleInfo>
<name type="personal">
<namePart type="given">Masahiro</namePart>
<namePart type="family">Kaneko</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Timothy</namePart>
<namePart type="family">Baldwin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolyn</namePart>
<namePart type="family">Rose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Violet</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-332-6</identifier>
</relatedItem>
<abstract>Large Language Models (LLMs) are trained on massive web-crawled corpora, often containing personal information, copyrighted text, and benchmark datasets. This inadvertent inclusion in the training dataset, known as data leakage, poses significant risks and could compromise the safety of LLM outputs. Despite its criticality, existing studies do not examine how leaked instances in the pre-training data influence LLMs’ output and detection capabilities. In this paper, we conduct an experimental survey to elucidate the relationship between data leakage in training datasets and its effects on the generation and detection by LLMs. Our experiments reveal that LLMs often generate outputs containing leaked information, even when there is little such data in the training dataset. Moreover, the fewer the leaked instances, the more difficult it becomes to detect such leakage. Finally, we demonstrate that enhancing leakage detection through few-shot learning can help mitigate the impact of the leakage rate in the training data on detection performance.</abstract>
<identifier type="citekey">kaneko-baldwin-2025-investigating</identifier>
<identifier type="doi">10.18653/v1/2025.emnlp-main.1201</identifier>
<location>
<url>https://aclanthology.org/2025.emnlp-main.1201/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>23545</start>
<end>23555</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Investigating How Pre-training Data Leakage Affects Models’ Reproduction and Detection Capabilities
%A Kaneko, Masahiro
%A Baldwin, Timothy
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-332-6
%F kaneko-baldwin-2025-investigating
%X Large Language Models (LLMs) are trained on massive web-crawled corpora, often containing personal information, copyrighted text, and benchmark datasets. This inadvertent inclusion in the training dataset, known as data leakage, poses significant risks and could compromise the safety of LLM outputs. Despite its criticality, existing studies do not examine how leaked instances in the pre-training data influence LLMs’ output and detection capabilities. In this paper, we conduct an experimental survey to elucidate the relationship between data leakage in training datasets and its effects on the generation and detection by LLMs. Our experiments reveal that LLMs often generate outputs containing leaked information, even when there is little such data in the training dataset. Moreover, the fewer the leaked instances, the more difficult it becomes to detect such leakage. Finally, we demonstrate that enhancing leakage detection through few-shot learning can help mitigate the impact of the leakage rate in the training data on detection performance.
%R 10.18653/v1/2025.emnlp-main.1201
%U https://aclanthology.org/2025.emnlp-main.1201/
%U https://doi.org/10.18653/v1/2025.emnlp-main.1201
%P 23545-23555
Markdown (Informal)
[Investigating How Pre-training Data Leakage Affects Models’ Reproduction and Detection Capabilities](https://aclanthology.org/2025.emnlp-main.1201/) (Kaneko & Baldwin, EMNLP 2025)
ACL