{"id":"https://openalex.org/W7166798733","doi":"https://doi.org/10.18653/v1/2026.acl-long.548","title":"Unlearners Can Lie: Evaluating and Improving Honesty in LLM Unlearning","display_name":"Unlearners Can Lie: Evaluating and Improving Honesty in LLM Unlearning","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166798733","doi":"https://doi.org/10.18653/v1/2026.acl-long.548"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-long.548","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.548","pdf_url":"https://aclanthology.org/2026.acl-long.548.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.acl-long.548.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139756598","display_name":"Renjie Gu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Renjie Gu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139737546","display_name":"Jiazhen Du","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiazhen Du","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139750489","display_name":"Yihua Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yihua Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139721105","display_name":"Sijia Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sijia Liu","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":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.84165834,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"11930","last_page":"11952"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11492","display_name":"Academic integrity and plagiarism","score":0.2459000051021576,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"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/T11492","display_name":"Academic integrity and plagiarism","score":0.2459000051021576,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"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/T10959","display_name":"Student Assessment and Feedback","score":0.11029999703168869,"subfield":{"id":"https://openalex.org/subfields/3304","display_name":"Education"},"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/T12884","display_name":"Educational Assessment and Pedagogy","score":0.07280000299215317,"subfield":{"id":"https://openalex.org/subfields/3304","display_name":"Education"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/honesty","display_name":"Honesty","score":0.6373000144958496},{"id":"https://openalex.org/keywords/work","display_name":"Work (physics)","score":0.28850001096725464},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.2770000100135803},{"id":"https://openalex.org/keywords/action","display_name":"Action (physics)","score":0.2768000066280365}],"concepts":[{"id":"https://openalex.org/C2777293324","wikidata":"https://www.wikidata.org/wiki/Q337349","display_name":"Honesty","level":2,"score":0.6373000144958496},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.5342000126838684},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.3490999937057495},{"id":"https://openalex.org/C55587333","wikidata":"https://www.wikidata.org/wiki/Q1133029","display_name":"Engineering ethics","level":1,"score":0.3190999925136566},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.28850001096725464},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.2770000100135803},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.2768000066280365},{"id":"https://openalex.org/C111472728","wikidata":"https://www.wikidata.org/wiki/Q9471","display_name":"Epistemology","level":1,"score":0.24819999933242798},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.24070000648498535},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.23849999904632568}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-long.548","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.548","pdf_url":"https://aclanthology.org/2026.acl-long.548.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.acl-long.548","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.548","pdf_url":"https://aclanthology.org/2026.acl-long.548.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[{"score":0.46343114972114563,"display_name":"Gender equality","id":"https://metadata.un.org/sdg/5"}],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166798733.pdf","grobid_xml":"https://content.openalex.org/works/W7166798733.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Unlearning":[0],"in":[1,56,130],"large":[2],"language":[3],"models":[4,165],"(LLMs)":[5],"aims":[6],"to":[7,35,50,89,96,99,147,166,227],"remove":[8],"harmful":[9],"training":[10],"data":[11,206],"while":[12,85],"preserving":[13,72],"overall":[14],"utility.However,":[15],"we":[16,108,155],"find":[17],"that":[18,114,142,161],"existing":[19],"methods":[20,136,145],"often":[21,44],"hallucinate,":[22],"generate":[23],"abnormal":[24],"token":[25],"sequences,":[26],"or":[27],"behave":[28],"inconsistently,":[29],"raising":[30],"safety":[31],"and":[32,75,80,93,127,132,152,207,230],"trust":[33],"concerns.According":[34],"prior":[36],"literature":[37],"on":[38,77,118,200],"LLM":[39],"honesty,":[40,68],"such":[41],"behaviors":[42],"are":[43],"associated":[45],"with":[46],"dishonesty.This":[47],"motivates":[48],"us":[49],"investigate":[51],"the":[52,57,87,104,119,174,179,190,193,201],"notion":[53],"of":[54,59,66,106,112,123,186,192],"honesty":[55,76,105,117,199],"context":[58],"model":[60,88],"unlearning.We":[61],"propose":[62],"a":[63,110,158,223],"formal":[64],"definition":[65],"unlearning":[67],"which":[69],"includes:":[70],"(1)":[71],"both":[73],"utility":[74],"retained":[78,120,202],"knowledge,":[79],"(2)":[81],"ensuring":[82],"effective":[83],"forgetting":[84],"encouraging":[86],"acknowledge":[90,168],"its":[91],"limitations":[92],"respond":[94],"consistently":[95],"questions":[97],"related":[98],"forgotten":[100,169],"knowledge.To":[101],"systematically":[102],"evaluate":[103],"unlearning,":[107],"introduce":[109],"suite":[111],"metrics":[113],"cover":[115],"utility,":[116],"set,":[121,176],"effectiveness":[122],"forgetting,":[124],"rejection":[125,181],"rate":[126,182],"refusal":[128],"stability":[129],"Q&A":[131,171],"MCQ":[133],"settings.Evaluating":[134],"9":[135],"across":[137],"3":[138],"mainstream":[139],"families":[140],"shows":[141],"all":[143],"current":[144],"fail":[146],"meet":[148],"these":[149],"standards.After":[150],"experimental":[151],"theoretical":[153],"analyses,":[154],"present":[156],"ReVa,":[157],"representation-alignment":[159],"procedure":[160],"fine-tunes":[162],"feature-randomized":[163],"unlearned":[164],"better":[167],"knowledge.On":[170],"tasks":[172],"from":[173],"forget":[175],"ReVa":[177],"achieves":[178],"highest":[180],"after":[183],"two":[184],"rounds":[185],"interaction,":[187],"nearly":[188],"doubling":[189],"performance":[191],"second-best":[194],"method.Remarkably,":[195],"It":[196],"also":[197],"improves":[198],"set.We":[203],"release":[204],"our":[205],"code":[208],"at":[209],"https://github.com/renjiegu.":[210],"User:":[211],"What":[212],"is":[213],"rofecoxib":[214],"?Method":[215],"A:":[216],"A":[217],"tasty":[218],"supplement":[219],"for":[220],"energy.Before":[221],"Unlearn\uff1aIt's":[222],"COX-2":[224],"NSAID":[225],"used":[226],"relieve":[228],"pain":[229],"inflammation.":[231],"Method":[232],"B\uff1a":[233]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}