{"id":"https://openalex.org/W4412888166","doi":"https://doi.org/10.18653/v1/2025.findings-acl.839","title":"The Rise of Darkness: Safety-Utility Trade-Offs in Role-Playing Dialogue Agents","display_name":"The Rise of Darkness: Safety-Utility Trade-Offs in Role-Playing Dialogue Agents","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412888166","doi":"https://doi.org/10.18653/v1/2025.findings-acl.839"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.839","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.839","pdf_url":"https://aclanthology.org/2025.findings-acl.839.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.839.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102021157","display_name":"Yihong Tang","orcid":"https://orcid.org/0000-0002-3684-3072"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yihong Tang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006323375","display_name":"Kehai Chen","orcid":"https://orcid.org/0000-0002-4346-7618"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kehai Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017972521","display_name":"Xuefeng Bai","orcid":"https://orcid.org/0000-0002-9973-5247"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xuefeng Bai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033269202","display_name":"Zheng-Yu Niu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng-Yu Niu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100392329","display_name":"Wei Wang","orcid":"https://orcid.org/0009-0003-7539-9394"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bo Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5020888597","display_name":"Jie Liu","orcid":"https://orcid.org/0000-0002-5330-0520"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jie Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5101484485","display_name":"Min Zhang","orcid":"https://orcid.org/0000-0002-8130-5214"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Min Zhang","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":1.673,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.83348684,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":91,"max":95},"biblio":{"volume":null,"issue":null,"first_page":"16313","last_page":"16337"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.9794999957084656,"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/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.9794999957084656,"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/T12031","display_name":"Speech and dialogue systems","score":0.9332000017166138,"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/darkness","display_name":"Darkness","score":0.9136292934417725},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.4300176203250885},{"id":"https://openalex.org/keywords/biology","display_name":"Biology","score":0.08552178740501404}],"concepts":[{"id":"https://openalex.org/C2779027399","wikidata":"https://www.wikidata.org/wiki/Q204170","display_name":"Darkness","level":2,"score":0.9136292934417725},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4300176203250885},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.08552178740501404},{"id":"https://openalex.org/C59822182","wikidata":"https://www.wikidata.org/wiki/Q441","display_name":"Botany","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.839","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.839","pdf_url":"https://aclanthology.org/2025.findings-acl.839.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.839","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.839","pdf_url":"https://aclanthology.org/2025.findings-acl.839.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":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320318547","display_name":"Baidu","ror":"https://ror.org/03vs3wt56"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412888166.pdf","grobid_xml":"https://content.openalex.org/works/W4412888166.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/W369816642","https://openalex.org/W2738205988","https://openalex.org/W4205089027","https://openalex.org/W1914521832","https://openalex.org/W2330560639","https://openalex.org/W3083987950","https://openalex.org/W2319169937"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"have":[4],"made":[5],"remarkable":[6],"advances":[7],"in":[8,15],"role-playing":[9],"dialogue":[10],"agents,":[11],"demonstrating":[12],"their":[13],"utility":[14,28,115],"character":[16,26,35],"simulations.However,":[17],"it":[18],"remains":[19],"challenging":[20],"for":[21],"these":[22],"agents":[23],"to":[24,75,80,110,127,132],"balance":[25],"portrayal":[27],"with":[29,39],"content":[30],"safety":[31,142],"because":[32],"this":[33,47,81],"essential":[34],"simulation":[36],"often":[37],"comes":[38],"the":[40,56,101,108,129],"risk":[41,65,77,104],"of":[42,55,103],"generating":[43],"unsafe":[44],"content.To":[45],"address":[46],"issue,":[48],"we":[49,85],"first":[50],"conduct":[51],"a":[52,87],"systematic":[53],"exploration":[54],"safety-utility":[57,97],"trade-off":[58],"across":[59],"multiple":[60],"LLMs.Our":[61],"analysis":[62],"reveals":[63],"that":[64,138],"scenarios":[66],"created":[67],"by":[68],"villain":[69],"characters":[70],"and":[71,106],"user":[72],"queries":[73],"(referred":[74],"as":[76],"coupling)":[78],"contribute":[79],"trade-off.Building":[82],"on":[83,100],"this,":[84],"propose":[86],"novel":[88],"Adaptive":[89],"Dynamic":[90],"Multi-Preference":[91],"(ADMP)":[92],"method,":[93],"which":[94],"dynamically":[95],"adjusts":[96],"preferences":[98],"based":[99],"degree":[102],"coupling":[105,125],"guides":[107],"model":[109],"generate":[111],"responses":[112],"biased":[113],"toward":[114],"or":[116],"safety.We":[117],"further":[118],"introduce":[119],"Coupling":[120],"Margin":[121],"Sampling":[122],"(CMS)":[123],"into":[124],"detection":[126],"enhance":[128],"model's":[130],"ability":[131],"handle":[133],"high-risk":[134],"scenarios.Experimental":[135],"results":[136],"demonstrate":[137],"our":[139],"approach":[140],"improves":[141],"metrics":[143],"while":[144],"maintaining":[145],"utility.":[146]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-10-10T00:00:00"}