{"id":"https://openalex.org/W4392086355","doi":"https://doi.org/10.48550/arxiv.2402.13459","title":"Learning to Poison Large Language Models for Downstream Manipulation","display_name":"Learning to Poison Large Language Models for Downstream Manipulation","publication_year":2024,"publication_date":"2024-02-21","ids":{"openalex":"https://openalex.org/W4392086355","doi":"https://doi.org/10.48550/arxiv.2402.13459"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2402.13459","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.13459","pdf_url":"https://arxiv.org/pdf/2402.13459","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2402.13459","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5060955674","display_name":"Xiangyu Zhou","orcid":"https://orcid.org/0009-0000-0776-8045","observed_orcids":["https://orcid.org/0009-0000-0776-8045"]},"institutions":[{"id":"https://openalex.org/I185443292","display_name":"Wayne State University","ror":"https://ror.org/01070mq45","country_code":"US","type":"education","lineage":["https://openalex.org/I185443292"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhou, Xiangyu","raw_affiliation_strings":["Department of Computer Science , Wayne State University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , Wayne State University","institution_ids":["https://openalex.org/I185443292"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5018415209","display_name":"Yao Qiang","orcid":"https://orcid.org/0000-0003-2995-3385","observed_orcids":["https://orcid.org/0000-0003-2995-3385"]},"institutions":[{"id":"https://openalex.org/I177721651","display_name":"Oakland University","ror":"https://ror.org/01ythxj32","country_code":"US","type":"education","lineage":["https://openalex.org/I177721651"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Qiang, Yao","raw_affiliation_strings":["Computer Science and Engineering Department , Oakland University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Computer Science and Engineering Department , Oakland University","institution_ids":["https://openalex.org/I177721651"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5093984391","display_name":"Saleh Zare Zade","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I185443292","display_name":"Wayne State University","ror":"https://ror.org/01070mq45","country_code":"US","type":"education","lineage":["https://openalex.org/I185443292"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zade, Saleh Zare","raw_affiliation_strings":["Department of Computer Science , Wayne State University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , Wayne State University","institution_ids":["https://openalex.org/I185443292"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5093984392","display_name":"Mohammad Amin Roshani","orcid":"https://orcid.org/0009-0006-2419-3101","observed_orcids":["https://orcid.org/0009-0006-2419-3101"]},"institutions":[{"id":"https://openalex.org/I185443292","display_name":"Wayne State University","ror":"https://ror.org/01070mq45","country_code":"US","type":"education","lineage":["https://openalex.org/I185443292"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Roshani, Mohammad Amin","raw_affiliation_strings":["Department of Computer Science , Wayne State University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , Wayne State University","institution_ids":["https://openalex.org/I185443292"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Khanduri, Prashant","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I185443292","display_name":"Wayne State University","ror":"https://ror.org/01070mq45","country_code":"US","type":"education","lineage":["https://openalex.org/I185443292"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Khanduri, Prashant","raw_affiliation_strings":["Department of Computer Science , Wayne State University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , Wayne State University","institution_ids":["https://openalex.org/I185443292"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033234764","display_name":"Douglas Zytko","orcid":"https://orcid.org/0000-0002-6854-5336","observed_orcids":["https://orcid.org/0000-0002-6854-5336"]},"institutions":[{"id":"https://openalex.org/I4210092198","display_name":"University of Michigan\u2013Flint","ror":"https://ror.org/01c3xc117","country_code":"US","type":"education","lineage":["https://openalex.org/I4210092198"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zytko, Douglas","raw_affiliation_strings":["College of Innovation and Technology , University of Michigan-Flint xiangyu , salehz , roshani ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"College of Innovation and Technology , University of Michigan-Flint xiangyu , salehz , roshani ,","institution_ids":["https://openalex.org/I4210092198"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5009256505","display_name":"Dongxiao Zhu","orcid":"https://orcid.org/0000-0002-0225-7817","observed_orcids":["https://orcid.org/0000-0002-0225-7817"]},"institutions":[{"id":"https://openalex.org/I185443292","display_name":"Wayne State University","ror":"https://ror.org/01070mq45","country_code":"US","type":"education","lineage":["https://openalex.org/I185443292"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhu, Dongxiao","raw_affiliation_strings":["Department of Computer Science , Wayne State University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science , Wayne State University","institution_ids":["https://openalex.org/I185443292"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":5,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":1.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10764","display_name":"Privacy-Preserving Technologies in Data","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11424","display_name":"Security and Verification in Computing","score":0.0,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/data-poisoning-attacks","display_name":"data poisoning","score":0.9279999732971191},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9729999899864197},{"id":"https://openalex.org/keywords/backdoor-triggers","display_name":"backdoor triggers","score":0.925000011920929},{"id":"https://openalex.org/keywords/supervised-fine-tuning","display_name":"supervised fine-tuning","score":0.8769999742507935},{"id":"https://openalex.org/keywords/adversarial-training-data","display_name":"adversarial training data","score":0.5460000038146973},{"id":"https://openalex.org/keywords/in-context-learning","display_name":"in-context learning","score":0.7289999723434448},{"id":"https://openalex.org/keywords/continuous-learning","display_name":"continuous learning","score":0.9279999732971191},{"id":"https://openalex.org/keywords/llm-security","display_name":"LLM security","score":0.7630000114440918}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6172720193862915}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2402.13459","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.13459","pdf_url":"https://arxiv.org/pdf/2402.13459","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2402.13459","doi":"https://doi.org/10.48550/arxiv.2402.13459","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2402.13459","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2402.13459","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2402.13459","pdf_url":"https://arxiv.org/pdf/2402.13459","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4392086355.pdf","grobid_xml":"https://content.openalex.org/works/W4392086355.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7162562020","https://openalex.org/W4416372094","https://openalex.org/W6930594876","https://openalex.org/W7154476108","https://openalex.org/W7125579043","https://openalex.org/W4417118608","https://openalex.org/W7103185145","https://openalex.org/W4408794189","https://openalex.org/W4399694499","https://openalex.org/W7138458336"],"abstract_inverted_index":{"The":[0],"advent":[1],"of":[2,81,142,161,166],"Large":[3],"Language":[4],"Models":[5],"(LLMs)":[6],"has":[7],"marked":[8],"significant":[9,155],"achievements":[10],"in":[11,46,114,149],"language":[12,95],"processing":[13],"and":[14,103,133,144,163],"reasoning":[15],"capabilities.":[16],"Despite":[17],"their":[18],"advancements,":[19],"LLMs":[20,47,143,162,168],"face":[21],"vulnerabilities":[22],"to":[23,36,56,73],"data":[24,35,52,126,170],"poisoning":[25,53,107,127,171],"attacks,":[26,128],"where":[27],"the":[28,58,140,147,154,164],"adversary":[29],"inserts":[30],"backdoor":[31,68],"triggers":[32,76],"into":[33],"training":[34],"manipulate":[37],"outputs.":[38,118],"This":[39],"work":[40,152],"further":[41,120],"identifies":[42],"additional":[43],"security":[44,156],"risks":[45,157],"by":[48,83],"designing":[49],"a":[50,65,110],"new":[51],"attack":[54],"tailored":[55],"exploit":[57],"supervised":[59],"fine-tuning":[60],"(SFT)":[61],"process.":[62],"We":[63,119],"propose":[64,121],"novel":[66],"gradient-guided":[67],"trigger":[69],"learning":[70,131,135],"(GBTL)":[71],"algorithm":[72],"identify":[74],"adversarial":[75],"efficiently,":[77],"ensuring":[78],"an":[79],"evasion":[80],"detection":[82],"conventional":[84],"defenses":[85],"while":[86],"maintaining":[87],"content":[88],"integrity.":[89],"Through":[90],"experimental":[91],"validation":[92],"across":[93],"various":[94,116],"model":[96],"tasks,":[97],"including":[98,129],"sentiment":[99],"analysis,":[100],"domain":[101],"generation,":[102],"question":[104],"answering,":[105],"our":[106],"strategy":[108],"demonstrates":[109],"high":[111],"success":[112],"rate":[113],"compromising":[115],"LLMs'":[117],"two":[122],"defense":[123],"strategies":[124],"against":[125,169],"in-context":[130],"(ICL)":[132],"continuous":[134],"(CL),":[136],"which":[137],"effectively":[138],"rectify":[139],"behavior":[141],"significantly":[145],"reduce":[146],"decline":[148],"performance.":[150],"Our":[151],"highlights":[153],"present":[158],"during":[159],"SFT":[160],"necessity":[165],"safeguarding":[167],"attacks.":[172]},"counts_by_year":[{"year":2025,"cited_by_count":4},{"year":2024,"cited_by_count":1}],"updated_date":"2026-10-16T09:59:22.235173","created_date":"2024-02-23T00:00:00"}