{"id":"https://openalex.org/W4416017657","doi":"https://doi.org/10.1145/3746252.3760929","title":"Uncovering the Persuasive Fingerprint of LLMs in Jailbreaking Attacks","display_name":"Uncovering the Persuasive Fingerprint of LLMs in Jailbreaking Attacks","publication_year":2025,"publication_date":"2025-11-08","ids":{"openalex":"https://openalex.org/W4416017657","doi":"https://doi.org/10.1145/3746252.3760929"},"language":null,"primary_location":{"id":"doi:10.1145/3746252.3760929","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3746252.3760929","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 34th ACM International Conference on Information and Knowledge Management","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5062961438","display_name":"Havva Alizadeh Noughabi","orcid":"https://orcid.org/0000-0002-9801-427X"},"institutions":[{"id":"https://openalex.org/I79817857","display_name":"University of Guelph","ror":"https://ror.org/01r7awg59","country_code":"CA","type":"education","lineage":["https://openalex.org/I79817857"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Havva Alizadeh Noughabi","raw_affiliation_strings":["Cyber Science Lab, University of Guelph, Guelph, ON, Canada"],"raw_orcid":"https://orcid.org/0000-0002-9801-427X","affiliations":[{"raw_affiliation_string":"Cyber Science Lab, University of Guelph, Guelph, ON, Canada","institution_ids":["https://openalex.org/I79817857"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5115046073","display_name":"Julien Serbanescu","orcid":"https://orcid.org/0009-0000-2305-0228"},"institutions":[{"id":"https://openalex.org/I79817857","display_name":"University of Guelph","ror":"https://ror.org/01r7awg59","country_code":"CA","type":"education","lineage":["https://openalex.org/I79817857"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Julien Serbanescu","raw_affiliation_strings":["Cyber Science Lab, University of Guelph, Guelph, ON, Canada"],"raw_orcid":"https://orcid.org/0009-0000-2305-0228","affiliations":[{"raw_affiliation_string":"Cyber Science Lab, University of Guelph, Guelph, ON, Canada","institution_ids":["https://openalex.org/I79817857"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5073867210","display_name":"Fattane Zarrinkalam","orcid":"https://orcid.org/0000-0002-2102-9190"},"institutions":[{"id":"https://openalex.org/I79817857","display_name":"University of Guelph","ror":"https://ror.org/01r7awg59","country_code":"CA","type":"education","lineage":["https://openalex.org/I79817857"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Fattane Zarrinkalam","raw_affiliation_strings":["College of Engineering, University of Guelph, Guelph, ON, Canada"],"raw_orcid":"https://orcid.org/0000-0002-2102-9190","affiliations":[{"raw_affiliation_string":"College of Engineering, University of Guelph, Guelph, ON, Canada","institution_ids":["https://openalex.org/I79817857"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5038019914","display_name":"Ali Dehghantanha","orcid":"https://orcid.org/0000-0002-9294-7554"},"institutions":[{"id":"https://openalex.org/I79817857","display_name":"University of Guelph","ror":"https://ror.org/01r7awg59","country_code":"CA","type":"education","lineage":["https://openalex.org/I79817857"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Ali Dehghantanha","raw_affiliation_strings":["Cyber Science Lab, University of Guelph, Guelph, ON, Canada"],"raw_orcid":"https://orcid.org/0000-0002-9294-7554","affiliations":[{"raw_affiliation_string":"Cyber Science Lab, University of Guelph, Guelph, ON, Canada","institution_ids":["https://openalex.org/I79817857"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I79817857"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.29381449,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"4608","last_page":"4612"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.3718999922275543,"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":0.3718999922275543,"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/T10734","display_name":"Information and Cyber Security","score":0.06700000166893005,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T12262","display_name":"Hate Speech and Cyberbullying Detection","score":0.04690000042319298,"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/persuasion","display_name":"Persuasion","score":0.7631000280380249},{"id":"https://openalex.org/keywords/adversarial-system","display_name":"Adversarial system","score":0.669700026512146},{"id":"https://openalex.org/keywords/readability","display_name":"Readability","score":0.6255999803543091},{"id":"https://openalex.org/keywords/fingerprint","display_name":"Fingerprint (computing)","score":0.4041999876499176},{"id":"https://openalex.org/keywords/interrogative","display_name":"Interrogative","score":0.39649999141693115},{"id":"https://openalex.org/keywords/sophistication","display_name":"Sophistication","score":0.39089998602867126}],"concepts":[{"id":"https://openalex.org/C2781310500","wikidata":"https://www.wikidata.org/wiki/Q1231428","display_name":"Persuasion","level":2,"score":0.7631000280380249},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.669700026512146},{"id":"https://openalex.org/C2778143727","wikidata":"https://www.wikidata.org/wiki/Q1820650","display_name":"Readability","level":2,"score":0.6255999803543091},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.49390000104904175},{"id":"https://openalex.org/C108827166","wikidata":"https://www.wikidata.org/wiki/Q175975","display_name":"Internet privacy","level":1,"score":0.4641000032424927},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.4544000029563904},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.41029998660087585},{"id":"https://openalex.org/C2777826928","wikidata":"https://www.wikidata.org/wiki/Q3745713","display_name":"Fingerprint (computing)","level":2,"score":0.4041999876499176},{"id":"https://openalex.org/C57098296","wikidata":"https://www.wikidata.org/wiki/Q12021746","display_name":"Interrogative","level":2,"score":0.39649999141693115},{"id":"https://openalex.org/C168725872","wikidata":"https://www.wikidata.org/wiki/Q991663","display_name":"Sophistication","level":2,"score":0.39089998602867126},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3774000108242035},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.34049999713897705},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.32839998602867126},{"id":"https://openalex.org/C2778223634","wikidata":"https://www.wikidata.org/wiki/Q224952","display_name":"Suspect","level":2,"score":0.311599999666214},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.31029999256134033},{"id":"https://openalex.org/C131158328","wikidata":"https://www.wikidata.org/wiki/Q1307337","display_name":"Social influence","level":2,"score":0.29170000553131104},{"id":"https://openalex.org/C115086926","wikidata":"https://www.wikidata.org/wiki/Q17004651","display_name":"Causal reasoning","level":3,"score":0.2849000096321106},{"id":"https://openalex.org/C2779732396","wikidata":"https://www.wikidata.org/wiki/Q2207288","display_name":"Craft","level":2,"score":0.2761000096797943},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.26969999074935913},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.2578999996185303}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3746252.3760929","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3746252.3760929","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 34th ACM International Conference on Information and Knowledge Management","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":7,"referenced_works":["https://openalex.org/W1176587911","https://openalex.org/W1970794815","https://openalex.org/W2905450671","https://openalex.org/W4402671039","https://openalex.org/W4403236745","https://openalex.org/W4405181600","https://openalex.org/W4410609100"],"related_works":[],"abstract_inverted_index":{"Despite":[0],"recent":[1],"advances,":[2],"Large":[3],"Language":[4],"Models":[5],"(LLMs)":[6],"remain":[7],"vulnerable":[8],"to":[9,39,51,74,105,142],"jailbreak":[10,124,144],"attacks":[11],"that":[12,45,64,92,120,133],"bypass":[13,137],"alignment":[14,81],"safeguards":[15],"and":[16,32,42,164],"elicit":[17],"harmful":[18],"outputs.":[19],"While":[20],"prior":[21],"research":[22,63],"has":[23,36],"proposed":[24],"various":[25],"attack":[26],"strategies":[27],"differing":[28],"in":[29,83,122,154],"human":[30],"readability":[31],"transferability,":[33],"little":[34],"attention":[35],"been":[37,95],"paid":[38],"the":[40,71,149,156],"linguistic":[41],"psychological":[43],"mechanisms":[44],"may":[46,101],"influence":[47],"a":[48],"model's":[49],"susceptibility":[50],"such":[52],"attacks.":[53],"In":[54],"this":[55],"paper,":[56],"we":[57,90,111],"examine":[58],"an":[59],"interdisciplinary":[60],"line":[61],"of":[62,68,79,151,159],"leverages":[65],"foundational":[66],"theories":[67],"persuasion":[69],"from":[70],"social":[72],"sciences":[73],"craft":[75],"adversarial":[76],"prompts":[77,106,135],"capable":[78],"circumventing":[80],"constraints":[82],"LLMs.":[84],"Drawing":[85],"on":[86,97],"well-established":[87],"persuasive":[88,108,118],"strategies,":[89],"hypothesize":[91],"LLMs,":[93],"having":[94],"trained":[96],"large-scale":[98],"human-generated":[99],"text,":[100],"respond":[102],"more":[103],"compliantly":[104],"with":[107],"structures.":[109],"Furthermore,":[110],"investigate":[112],"whether":[113],"LLMs":[114,131],"themselves":[115],"exhibit":[116],"distinct":[117],"fingerprints":[119],"emerge":[121],"their":[123,140],"responses.":[125],"Empirical":[126],"evaluations":[127],"across":[128],"multiple":[129],"aligned":[130],"reveal":[132],"persuasion-aware":[134],"significantly":[136],"safeguards,":[138],"demonstrating":[139],"potential":[141],"induce":[143],"behaviors.":[145],"This":[146],"work":[147],"underscores":[148],"importance":[150],"cross-disciplinary":[152],"insight":[153],"addressing":[155],"evolving":[157],"challenges":[158],"LLM":[160],"safety.":[161],"The":[162],"code":[163],"data":[165],"are":[166],"available.":[167],"https://github.com/CyberScienceLab/Our-Papers/tree/main/PersuasiveJailbreaking/.":[168]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-11-08T00:00:00"}