Model for Peanuts: Hijacking ML Models without Training Access is Possible
The massive deployment of Machine Learning (ML) models has been accompanied by the emergence of several attacks that threaten their trustworthiness and raise ethical and societal concerns such as invasion of privacy, discrimination risks, and lack of accountability. Model hijacking is one of these a...
Gespeichert in:
Veröffentlicht in: | arXiv.org 2024-06 |
---|---|
Hauptverfasser: | , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
container_end_page | |
---|---|
container_issue | |
container_start_page | |
container_title | arXiv.org |
container_volume | |
creator | Ghorbel, Mahmoud Bouzidi, Halima Bilasco, Ioan Marius Alouani, Ihsen |
description | The massive deployment of Machine Learning (ML) models has been accompanied by the emergence of several attacks that threaten their trustworthiness and raise ethical and societal concerns such as invasion of privacy, discrimination risks, and lack of accountability. Model hijacking is one of these attacks, where the adversary aims to hijack a victim model to execute a different task than its original one. Model hijacking can cause accountability and security risks since a hijacked model owner can be framed for having their model offering illegal or unethical services. Prior state-of-the-art works consider model hijacking as a training time attack, whereby an adversary requires access to the ML model training to execute their attack. In this paper, we consider a stronger threat model where the attacker has no access to the training phase of the victim model. Our intuition is that ML models, typically over-parameterized, might (unintentionally) learn more than the intended task for they are trained. We propose a simple approach for model hijacking at inference time named SnatchML to classify unknown input samples using distance measures in the latent space of the victim model to previously known samples associated with the hijacking task classes. SnatchML empirically shows that benign pre-trained models can execute tasks that are semantically related to the initial task. Surprisingly, this can be true even for hijacking tasks unrelated to the original task. We also explore different methods to mitigate this risk. We first propose a novel approach we call meta-unlearning, designed to help the model unlearn a potentially malicious task while training on the original task dataset. We also provide insights on over-parameterization as one possible inherent factor that makes model hijacking easier, and we accordingly propose a compression-based countermeasure against this attack. |
format | Article |
fullrecord | <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_3064739683</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>3064739683</sourcerecordid><originalsourceid>FETCH-proquest_journals_30647396833</originalsourceid><addsrcrecordid>eNqNyksKwjAUQNEgCBbtHh44LsSkP52JKEUsdNB5SWuqqSXRvAS37wcX4OgOzp2QgHG-ivKYsRkJEQdKKUszliQ8IMfSnOUIvbFQSaG9ww0UahDdTekLlCf4OsJTuavxDmorlP7QtuskIiiEyiCqdpQLMu3FiDL8dU6Wh329K6K7NQ8v0TWD8Va_qeE0jTO-TnPO_7teWZ08Bw</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>3064739683</pqid></control><display><type>article</type><title>Model for Peanuts: Hijacking ML Models without Training Access is Possible</title><source>Free E- Journals</source><creator>Ghorbel, Mahmoud ; Bouzidi, Halima ; Bilasco, Ioan Marius ; Alouani, Ihsen</creator><creatorcontrib>Ghorbel, Mahmoud ; Bouzidi, Halima ; Bilasco, Ioan Marius ; Alouani, Ihsen</creatorcontrib><description>The massive deployment of Machine Learning (ML) models has been accompanied by the emergence of several attacks that threaten their trustworthiness and raise ethical and societal concerns such as invasion of privacy, discrimination risks, and lack of accountability. Model hijacking is one of these attacks, where the adversary aims to hijack a victim model to execute a different task than its original one. Model hijacking can cause accountability and security risks since a hijacked model owner can be framed for having their model offering illegal or unethical services. Prior state-of-the-art works consider model hijacking as a training time attack, whereby an adversary requires access to the ML model training to execute their attack. In this paper, we consider a stronger threat model where the attacker has no access to the training phase of the victim model. Our intuition is that ML models, typically over-parameterized, might (unintentionally) learn more than the intended task for they are trained. We propose a simple approach for model hijacking at inference time named SnatchML to classify unknown input samples using distance measures in the latent space of the victim model to previously known samples associated with the hijacking task classes. SnatchML empirically shows that benign pre-trained models can execute tasks that are semantically related to the initial task. Surprisingly, this can be true even for hijacking tasks unrelated to the original task. We also explore different methods to mitigate this risk. We first propose a novel approach we call meta-unlearning, designed to help the model unlearn a potentially malicious task while training on the original task dataset. We also provide insights on over-parameterization as one possible inherent factor that makes model hijacking easier, and we accordingly propose a compression-based countermeasure against this attack.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Accountability ; Machine learning ; Parameterization</subject><ispartof>arXiv.org, 2024-06</ispartof><rights>2024. This work is published under http://arxiv.org/licenses/nonexclusive-distrib/1.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>780,784</link.rule.ids></links><search><creatorcontrib>Ghorbel, Mahmoud</creatorcontrib><creatorcontrib>Bouzidi, Halima</creatorcontrib><creatorcontrib>Bilasco, Ioan Marius</creatorcontrib><creatorcontrib>Alouani, Ihsen</creatorcontrib><title>Model for Peanuts: Hijacking ML Models without Training Access is Possible</title><title>arXiv.org</title><description>The massive deployment of Machine Learning (ML) models has been accompanied by the emergence of several attacks that threaten their trustworthiness and raise ethical and societal concerns such as invasion of privacy, discrimination risks, and lack of accountability. Model hijacking is one of these attacks, where the adversary aims to hijack a victim model to execute a different task than its original one. Model hijacking can cause accountability and security risks since a hijacked model owner can be framed for having their model offering illegal or unethical services. Prior state-of-the-art works consider model hijacking as a training time attack, whereby an adversary requires access to the ML model training to execute their attack. In this paper, we consider a stronger threat model where the attacker has no access to the training phase of the victim model. Our intuition is that ML models, typically over-parameterized, might (unintentionally) learn more than the intended task for they are trained. We propose a simple approach for model hijacking at inference time named SnatchML to classify unknown input samples using distance measures in the latent space of the victim model to previously known samples associated with the hijacking task classes. SnatchML empirically shows that benign pre-trained models can execute tasks that are semantically related to the initial task. Surprisingly, this can be true even for hijacking tasks unrelated to the original task. We also explore different methods to mitigate this risk. We first propose a novel approach we call meta-unlearning, designed to help the model unlearn a potentially malicious task while training on the original task dataset. We also provide insights on over-parameterization as one possible inherent factor that makes model hijacking easier, and we accordingly propose a compression-based countermeasure against this attack.</description><subject>Accountability</subject><subject>Machine learning</subject><subject>Parameterization</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>ABUWG</sourceid><sourceid>AFKRA</sourceid><sourceid>AZQEC</sourceid><sourceid>BENPR</sourceid><sourceid>CCPQU</sourceid><sourceid>DWQXO</sourceid><recordid>eNqNyksKwjAUQNEgCBbtHh44LsSkP52JKEUsdNB5SWuqqSXRvAS37wcX4OgOzp2QgHG-ivKYsRkJEQdKKUszliQ8IMfSnOUIvbFQSaG9ww0UahDdTekLlCf4OsJTuavxDmorlP7QtuskIiiEyiCqdpQLMu3FiDL8dU6Wh329K6K7NQ8v0TWD8Va_qeE0jTO-TnPO_7teWZ08Bw</recordid><startdate>20240603</startdate><enddate>20240603</enddate><creator>Ghorbel, Mahmoud</creator><creator>Bouzidi, Halima</creator><creator>Bilasco, Ioan Marius</creator><creator>Alouani, Ihsen</creator><general>Cornell University Library, arXiv.org</general><scope>8FE</scope><scope>8FG</scope><scope>ABJCF</scope><scope>ABUWG</scope><scope>AFKRA</scope><scope>AZQEC</scope><scope>BENPR</scope><scope>BGLVJ</scope><scope>CCPQU</scope><scope>DWQXO</scope><scope>HCIFZ</scope><scope>L6V</scope><scope>M7S</scope><scope>PIMPY</scope><scope>PQEST</scope><scope>PQQKQ</scope><scope>PQUKI</scope><scope>PRINS</scope><scope>PTHSS</scope></search><sort><creationdate>20240603</creationdate><title>Model for Peanuts: Hijacking ML Models without Training Access is Possible</title><author>Ghorbel, Mahmoud ; Bouzidi, Halima ; Bilasco, Ioan Marius ; Alouani, Ihsen</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_30647396833</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Accountability</topic><topic>Machine learning</topic><topic>Parameterization</topic><toplevel>online_resources</toplevel><creatorcontrib>Ghorbel, Mahmoud</creatorcontrib><creatorcontrib>Bouzidi, Halima</creatorcontrib><creatorcontrib>Bilasco, Ioan Marius</creatorcontrib><creatorcontrib>Alouani, Ihsen</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science & Engineering Collection</collection><collection>ProQuest Central (Alumni Edition)</collection><collection>ProQuest Central UK/Ireland</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central Korea</collection><collection>SciTech Premium Collection</collection><collection>ProQuest Engineering Collection</collection><collection>Engineering Database</collection><collection>Publicly Available Content Database</collection><collection>ProQuest One Academic Eastern Edition (DO NOT USE)</collection><collection>ProQuest One Academic</collection><collection>ProQuest One Academic UKI Edition</collection><collection>ProQuest Central China</collection><collection>Engineering Collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Ghorbel, Mahmoud</au><au>Bouzidi, Halima</au><au>Bilasco, Ioan Marius</au><au>Alouani, Ihsen</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Model for Peanuts: Hijacking ML Models without Training Access is Possible</atitle><jtitle>arXiv.org</jtitle><date>2024-06-03</date><risdate>2024</risdate><eissn>2331-8422</eissn><abstract>The massive deployment of Machine Learning (ML) models has been accompanied by the emergence of several attacks that threaten their trustworthiness and raise ethical and societal concerns such as invasion of privacy, discrimination risks, and lack of accountability. Model hijacking is one of these attacks, where the adversary aims to hijack a victim model to execute a different task than its original one. Model hijacking can cause accountability and security risks since a hijacked model owner can be framed for having their model offering illegal or unethical services. Prior state-of-the-art works consider model hijacking as a training time attack, whereby an adversary requires access to the ML model training to execute their attack. In this paper, we consider a stronger threat model where the attacker has no access to the training phase of the victim model. Our intuition is that ML models, typically over-parameterized, might (unintentionally) learn more than the intended task for they are trained. We propose a simple approach for model hijacking at inference time named SnatchML to classify unknown input samples using distance measures in the latent space of the victim model to previously known samples associated with the hijacking task classes. SnatchML empirically shows that benign pre-trained models can execute tasks that are semantically related to the initial task. Surprisingly, this can be true even for hijacking tasks unrelated to the original task. We also explore different methods to mitigate this risk. We first propose a novel approach we call meta-unlearning, designed to help the model unlearn a potentially malicious task while training on the original task dataset. We also provide insights on over-parameterization as one possible inherent factor that makes model hijacking easier, and we accordingly propose a compression-based countermeasure against this attack.</abstract><cop>Ithaca</cop><pub>Cornell University Library, arXiv.org</pub><oa>free_for_read</oa></addata></record> |
fulltext | fulltext |
identifier | EISSN: 2331-8422 |
ispartof | arXiv.org, 2024-06 |
issn | 2331-8422 |
language | eng |
recordid | cdi_proquest_journals_3064739683 |
source | Free E- Journals |
subjects | Accountability Machine learning Parameterization |
title | Model for Peanuts: Hijacking ML Models without Training Access is Possible |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-06T12%3A25%3A47IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=document&rft.atitle=Model%20for%20Peanuts:%20Hijacking%20ML%20Models%20without%20Training%20Access%20is%20Possible&rft.jtitle=arXiv.org&rft.au=Ghorbel,%20Mahmoud&rft.date=2024-06-03&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E3064739683%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=3064739683&rft_id=info:pmid/&rfr_iscdi=true |