@inproceedings{mcgovern-etal-2025-large,
title = "Your Large Language Models are Leaving Fingerprints",
author = "McGovern, Hope Elizabeth and
Stureborg, Rickard and
Suhara, Yoshi and
Alikaniotis, Dimitris",
editor = "Alam, Firoj and
Nakov, Preslav and
Habash, Nizar and
Gurevych, Iryna and
Chowdhury, Shammur and
Shelmanov, Artem and
Wang, Yuxia and
Artemova, Ekaterina and
Kutlu, Mucahid and
Mikros, George",
booktitle = "Proceedings of the 1stWorkshop on GenAI Content Detection (GenAIDetect)",
month = jan,
year = "2025",
address = "Abu Dhabi, UAE",
publisher = "International Conference on Computational Linguistics",
url = "https://aclanthology.org/2025.genaidetect-1.6/",
pages = "85--95",
abstract = "It has been shown that fine-tuned transformers and other supervised detectors are effective for distinguishing between human and machine-generated texts in non-adversarial settings, but we find that even simple classifiers on top of n-gram and part-of-speech features can achieve very robust performance on both in- and out-of-domain data. To understand how this is possible, we analyze machine-generated output text in four datasets, finding that LLMs possess unique fingerprints that manifest as slight differences in the frequency of certain lexical and morphosyntactic features. We show how to visualize such fingerprints, describe how they can be used to detect machine-generated text and find that they are even robust across text domains. We find that fingerprints are often persistent across models in the same model family (e.g. 13B parameter LLaMA`s fingerprint is similar to that of 65B parameter LLaMA) and that while a detector trained on text from one model can easily recognize text generated by a model in the same family, it struggles to detect text generated by an unrelated model."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="mcgovern-etal-2025-large">
<titleInfo>
<title>Your Large Language Models are Leaving Fingerprints</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hope</namePart>
<namePart type="given">Elizabeth</namePart>
<namePart type="family">McGovern</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rickard</namePart>
<namePart type="family">Stureborg</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yoshi</namePart>
<namePart type="family">Suhara</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dimitris</namePart>
<namePart type="family">Alikaniotis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-01</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 1stWorkshop on GenAI Content Detection (GenAIDetect)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Firoj</namePart>
<namePart type="family">Alam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Preslav</namePart>
<namePart type="family">Nakov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nizar</namePart>
<namePart type="family">Habash</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Iryna</namePart>
<namePart type="family">Gurevych</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shammur</namePart>
<namePart type="family">Chowdhury</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Artem</namePart>
<namePart type="family">Shelmanov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuxia</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Artemova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mucahid</namePart>
<namePart type="family">Kutlu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">George</namePart>
<namePart type="family">Mikros</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>International Conference on Computational Linguistics</publisher>
<place>
<placeTerm type="text">Abu Dhabi, UAE</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>It has been shown that fine-tuned transformers and other supervised detectors are effective for distinguishing between human and machine-generated texts in non-adversarial settings, but we find that even simple classifiers on top of n-gram and part-of-speech features can achieve very robust performance on both in- and out-of-domain data. To understand how this is possible, we analyze machine-generated output text in four datasets, finding that LLMs possess unique fingerprints that manifest as slight differences in the frequency of certain lexical and morphosyntactic features. We show how to visualize such fingerprints, describe how they can be used to detect machine-generated text and find that they are even robust across text domains. We find that fingerprints are often persistent across models in the same model family (e.g. 13B parameter LLaMA‘s fingerprint is similar to that of 65B parameter LLaMA) and that while a detector trained on text from one model can easily recognize text generated by a model in the same family, it struggles to detect text generated by an unrelated model.</abstract>
<identifier type="citekey">mcgovern-etal-2025-large</identifier>
<location>
<url>https://aclanthology.org/2025.genaidetect-1.6/</url>
</location>
<part>
<date>2025-01</date>
<extent unit="page">
<start>85</start>
<end>95</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Your Large Language Models are Leaving Fingerprints
%A McGovern, Hope Elizabeth
%A Stureborg, Rickard
%A Suhara, Yoshi
%A Alikaniotis, Dimitris
%Y Alam, Firoj
%Y Nakov, Preslav
%Y Habash, Nizar
%Y Gurevych, Iryna
%Y Chowdhury, Shammur
%Y Shelmanov, Artem
%Y Wang, Yuxia
%Y Artemova, Ekaterina
%Y Kutlu, Mucahid
%Y Mikros, George
%S Proceedings of the 1stWorkshop on GenAI Content Detection (GenAIDetect)
%D 2025
%8 January
%I International Conference on Computational Linguistics
%C Abu Dhabi, UAE
%F mcgovern-etal-2025-large
%X It has been shown that fine-tuned transformers and other supervised detectors are effective for distinguishing between human and machine-generated texts in non-adversarial settings, but we find that even simple classifiers on top of n-gram and part-of-speech features can achieve very robust performance on both in- and out-of-domain data. To understand how this is possible, we analyze machine-generated output text in four datasets, finding that LLMs possess unique fingerprints that manifest as slight differences in the frequency of certain lexical and morphosyntactic features. We show how to visualize such fingerprints, describe how they can be used to detect machine-generated text and find that they are even robust across text domains. We find that fingerprints are often persistent across models in the same model family (e.g. 13B parameter LLaMA‘s fingerprint is similar to that of 65B parameter LLaMA) and that while a detector trained on text from one model can easily recognize text generated by a model in the same family, it struggles to detect text generated by an unrelated model.
%U https://aclanthology.org/2025.genaidetect-1.6/
%P 85-95
Markdown (Informal)
[Your Large Language Models are Leaving Fingerprints](https://aclanthology.org/2025.genaidetect-1.6/) (McGovern et al., GenAIDetect 2025)
ACL
- Hope Elizabeth McGovern, Rickard Stureborg, Yoshi Suhara, and Dimitris Alikaniotis. 2025. Your Large Language Models are Leaving Fingerprints. In Proceedings of the 1stWorkshop on GenAI Content Detection (GenAIDetect), pages 85–95, Abu Dhabi, UAE. International Conference on Computational Linguistics.