{"id":"https://openalex.org/W4412888092","doi":"https://doi.org/10.18653/v1/2025.findings-acl.848","title":"TUBA: Cross-Lingual Transferability of Backdoor Attacks in LLMs with Instruction Tuning","display_name":"TUBA: Cross-Lingual Transferability of Backdoor Attacks in LLMs with Instruction Tuning","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412888092","doi":"https://doi.org/10.18653/v1/2025.findings-acl.848"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.848","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.848","pdf_url":"https://aclanthology.org/2025.findings-acl.848.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.848.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5060320196","display_name":"Xuanli He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xuanli He","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100384686","display_name":"Jun Wang","orcid":"https://orcid.org/0000-0002-1305-5735"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jun Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5036737229","display_name":"Qiongkai Xu","orcid":"https://orcid.org/0000-0003-3312-6825"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qiongkai Xu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5096381037","display_name":"Pasquale Minervini","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pasquale Minervini","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049427071","display_name":"Pontus Stenetorp","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pontus Stenetorp","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5078824132","display_name":"Benjamin I. P. Rubinstein","orcid":"https://orcid.org/0000-0002-2947-6980"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Benjamin I. P. Rubinstein","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5078530959","display_name":"Trevor Cohn","orcid":"https://orcid.org/0000-0003-4363-1673"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Trevor Cohn","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":3,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"16504","last_page":"16544"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10927","display_name":"Access Control and Trust","score":0.9699000120162964,"subfield":{"id":"https://openalex.org/subfields/3312","display_name":"Sociology and Political Science"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T10927","display_name":"Access Control and Trust","score":0.9699000120162964,"subfield":{"id":"https://openalex.org/subfields/3312","display_name":"Sociology and Political Science"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9082000255584717,"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/backdoor","display_name":"Backdoor","score":0.9763401746749878},{"id":"https://openalex.org/keywords/transferability","display_name":"Transferability","score":0.8093801736831665},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.577807605266571},{"id":"https://openalex.org/keywords/computer-security","display_name":"Computer security","score":0.45874840021133423},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.18839555978775024}],"concepts":[{"id":"https://openalex.org/C2781045450","wikidata":"https://www.wikidata.org/wiki/Q254569","display_name":"Backdoor","level":2,"score":0.9763401746749878},{"id":"https://openalex.org/C61272859","wikidata":"https://www.wikidata.org/wiki/Q7834031","display_name":"Transferability","level":3,"score":0.8093801736831665},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.577807605266571},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.45874840021133423},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.18839555978775024},{"id":"https://openalex.org/C140331021","wikidata":"https://www.wikidata.org/wiki/Q1868104","display_name":"Logit","level":2,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.848","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.848","pdf_url":"https://aclanthology.org/2025.findings-acl.848.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.848","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.848","pdf_url":"https://aclanthology.org/2025.findings-acl.848.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[{"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4","score":0.6399999856948853}],"awards":[{"id":"https://openalex.org/G5730701950","display_name":null,"funder_award_id":"EP/W002876/1","funder_id":"https://openalex.org/F4320334627","funder_display_name":"Engineering and Physical Sciences Research Council"}],"funders":[{"id":"https://openalex.org/F4320307791","display_name":"Cisco Systems","ror":"https://ror.org/03yt1ez60"},{"id":"https://openalex.org/F4320307894","display_name":"Accenture","ror":"https://ror.org/013g16z83"},{"id":"https://openalex.org/F4320334627","display_name":"Engineering and Physical Sciences Research Council","ror":"https://ror.org/0439y7842"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412888092.pdf","grobid_xml":"https://content.openalex.org/works/W4412888092.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W4320031223","https://openalex.org/W4200629851","https://openalex.org/W4281902577","https://openalex.org/W4309417370","https://openalex.org/W4292107232","https://openalex.org/W3009072493","https://openalex.org/W4401407399"],"abstract_inverted_index":{"The":[0],"implications":[1],"of":[2,48,120,182],"backdoor":[3,49,61,139,165,224],"attacks":[4,16,50,62],"on":[5,51,59,148],"English-centric":[6],"large":[7],"language":[8],"models":[9,103,132],"(LLMs)":[10],"have":[11],"been":[12],"widely":[13],"examined":[14],"-such":[15],"can":[17,78],"be":[18],"achieved":[19],"by":[20,222],"embedding":[21],"malicious":[22,33],"behaviors":[23],"during":[24],"training":[25],"and":[26,43,106,156,184,205],"activated":[27],"under":[28],"specific":[29],"conditions":[30],"that":[31,96,129],"trigger":[32],"outputs.Despite":[34],"the":[35,46,70,80,164,187,207,218],"increasing":[36],"support":[37],"for":[38,73,82,210],"multilingual":[39,64,203],"capabilities":[40],"in":[41,102,115,169,202],"open-source":[42],"proprietary":[44],"LLMs,":[45,65],"impact":[47],"these":[52],"systems":[53],"remains":[54,190],"largely":[55],"under-explored.Our":[56],"research":[57],"focuses":[58],"cross-lingual":[60,138,170,223],"against":[63],"particularly":[66],"investigating":[67],"how":[68],"poisoning":[69],"instructiontuning":[71,85],"data":[72,86],"one":[74],"or":[75],"two":[76],"languages":[77,83,122],"affect":[79],"outputs":[81],"whose":[84],"were":[87],"not":[88],"poisoned.Despite":[89],"its":[90],"simplicity,":[91],"our":[92,97,158],"empirical":[93],"analysis":[94],"reveals":[95],"method":[98],"exhibits":[99],"remarkable":[100],"efficacy":[101],"like":[104],"BLOOM":[105],"GPT-4o,":[107],"with":[108],"high":[109],"attack":[110,179,189],"success":[111,180],"rates,":[112],"surpassing":[113],"90%":[114],"more":[116,130,211],"than":[117],"7":[118],"out":[119],"12":[121],"across":[123,173],"various":[124],"scenarios.Our":[125],"findings":[126,197],"also":[127,142],"indicate":[128],"powerful":[131],"show":[133],"increased":[134],"susceptibility":[135],"to":[136,144,216],"transferable":[137],"attacks,":[140],"which":[141],"applies":[143],"LLMs":[145,204],"predominantly":[146],"pre-trained":[147],"English/Chinese":[149],"data,":[150],"such":[151],"as":[152],"Llama2,":[153],"Llama3,":[154],"Qwen2.5,":[155],"Gemma.Moreover,":[157],"experiments":[159],"demonstrate":[160],"1)":[161],"High":[162],"Transferability:":[163],"mechanism":[166],"operates":[167],"successfully":[168],"response":[171],"scenarios":[172],"26":[174],"languages,":[175],"achieving":[176],"an":[177],"average":[178],"rate":[181],"99%,":[183],"2)":[185],"Robustness:":[186],"proposed":[188],"effective":[191],"even":[192],"after":[193],"defenses":[194],"are":[195],"applied.These":[196],"expose":[198],"critical":[199],"security":[200],"vulnerabilities":[201],"highlight":[206],"urgent":[208],"need":[209],"robust,":[212],"targeted":[213],"defense":[214],"strategies":[215],"address":[217],"unique":[219],"challenges":[220],"posed":[221],"transfer.":[225]},"counts_by_year":[{"year":2026,"cited_by_count":1},{"year":2025,"cited_by_count":2}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}