{"id":"https://openalex.org/W7140141059","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.53","title":"The Unintended Trade-off of AI Alignment: Balancing Hallucination Mitigation and Safety in LLMs","display_name":"The Unintended Trade-off of AI Alignment: Balancing Hallucination Mitigation and Safety in LLMs","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7140141059","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.53"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2026.findings-eacl.53","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.53","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.53","pdf_url":"https://aclanthology.org/2026.findings-eacl.53.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://aclanthology.org/2026.findings-eacl.53.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5103852720","display_name":"Omar Mahmoud","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I149704539","display_name":"Deakin University","ror":"https://ror.org/02czsnj07","country_code":"AU","type":"education","lineage":["https://openalex.org/I149704539"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Omar Mahmoud","raw_affiliation_strings":["Applied Artificial Intelligence Initiative , Deakin University , Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Applied Artificial Intelligence Initiative , Deakin University , Australia","institution_ids":["https://openalex.org/I149704539"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120365961","display_name":"Ali Khalil","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I149704539","display_name":"Deakin University","ror":"https://ror.org/02czsnj07","country_code":"AU","type":"education","lineage":["https://openalex.org/I149704539"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Ali Khalil","raw_affiliation_strings":["Applied Artificial Intelligence Initiative , Deakin University , Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Applied Artificial Intelligence Initiative , Deakin University , Australia","institution_ids":["https://openalex.org/I149704539"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5043916805","display_name":"Thommen George Karimpanal","orcid":"https://orcid.org/0000-0001-8918-3314","observed_orcids":["https://orcid.org/0000-0001-8918-3314"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Thommen George Karimpanal","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5019310532","display_name":"Buddhika Laknath Semage","orcid":"https://orcid.org/0000-0002-2792-492X","observed_orcids":["https://orcid.org/0000-0002-2792-492X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Buddhika Laknath Semage","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5024215125","display_name":"Santu Rana","orcid":"https://orcid.org/0000-0003-2247-850X","observed_orcids":["https://orcid.org/0000-0003-2247-850X"]},"institutions":[{"id":"https://openalex.org/I149704539","display_name":"Deakin University","ror":"https://ror.org/02czsnj07","country_code":"AU","type":"education","lineage":["https://openalex.org/I149704539"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Santu Rana","raw_affiliation_strings":["Applied Artificial Intelligence Initiative , Deakin University , Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Applied Artificial Intelligence Initiative , Deakin University , Australia","institution_ids":["https://openalex.org/I149704539"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":18.841,"has_fulltext":true,"cited_by_count":2,"citation_normalized_percentile":{"value":0.98724442,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":96,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"1017","last_page":"1037"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.36719998717308044,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.36719998717308044,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.24459999799728394,"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/T10028","display_name":"Topic Modeling","score":0.20340000092983246,"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/hallucination-mitigation","display_name":"hallucination mitigation","score":0.8069999814033508},{"id":"https://openalex.org/keywords/ai-alignment","display_name":"AI alignment","score":0.625},{"id":"https://openalex.org/keywords/honesty","display_name":"honesty","score":0.4429999887943268},{"id":"https://openalex.org/keywords/refusal-behavior","display_name":"refusal behavior","score":0.8930000066757202},{"id":"https://openalex.org/keywords/factual-accuracy","display_name":"factual accuracy","score":0.7429999709129333},{"id":"https://openalex.org/keywords/safety-alignment","display_name":"safety alignment","score":0.6710000038146973},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7200000286102295},{"id":"https://openalex.org/keywords/sparse-autoencoders","display_name":"sparse autoencoders","score":0.9100000262260437},{"id":"https://openalex.org/keywords/commonsense-reasoning","display_name":"commonsense reasoning","score":0.8130000233650208}],"concepts":[{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.42660000920295715},{"id":"https://openalex.org/C73484699","wikidata":"https://www.wikidata.org/wiki/Q161733","display_name":"Criminology","level":1,"score":0.41350001096725464},{"id":"https://openalex.org/C2776889888","wikidata":"https://www.wikidata.org/wiki/Q1135789","display_name":"Unintended consequences","level":2,"score":0.3433000147342682},{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.3416999876499176},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.33550000190734863},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.3294000029563904},{"id":"https://openalex.org/C95457728","wikidata":"https://www.wikidata.org/wiki/Q309","display_name":"History","level":0,"score":0.32580000162124634},{"id":"https://openalex.org/C39549134","wikidata":"https://www.wikidata.org/wiki/Q133080","display_name":"Public relations","level":1,"score":0.32409998774528503},{"id":"https://openalex.org/C95713431","wikidata":"https://www.wikidata.org/wiki/Q631425","display_name":"Vulnerability (computing)","level":2,"score":0.28220000863075256},{"id":"https://openalex.org/C32896092","wikidata":"https://www.wikidata.org/wiki/Q189447","display_name":"Risk management","level":2,"score":0.2797999978065491},{"id":"https://openalex.org/C50522688","wikidata":"https://www.wikidata.org/wiki/Q189833","display_name":"Economic growth","level":1,"score":0.2612000107765198},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.25949999690055847},{"id":"https://openalex.org/C144133560","wikidata":"https://www.wikidata.org/wiki/Q4830453","display_name":"Business","level":0,"score":0.25619998574256897},{"id":"https://openalex.org/C144024400","wikidata":"https://www.wikidata.org/wiki/Q21201","display_name":"Sociology","level":0,"score":0.25540000200271606},{"id":"https://openalex.org/C91375879","wikidata":"https://www.wikidata.org/wiki/Q15473274","display_name":"Environmental planning","level":1,"score":0.25130000710487366},{"id":"https://openalex.org/C12174686","wikidata":"https://www.wikidata.org/wiki/Q1058438","display_name":"Risk assessment","level":2,"score":0.25040000677108765}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2026.findings-eacl.53","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.53","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.53","pdf_url":"https://aclanthology.org/2026.findings-eacl.53.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2510.07775","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2510.07775","pdf_url":"https://arxiv.org/pdf/2510.07775","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-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2510.07775","doi":"https://doi.org/10.48550/arxiv.2510.07775","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2510.07775","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-eacl.53","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.53","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.53","pdf_url":"https://aclanthology.org/2026.findings-eacl.53.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140141059.pdf","grobid_xml":"https://content.openalex.org/works/W7140141059.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7084062577","https://openalex.org/W4415795809","https://openalex.org/W4412888863","https://openalex.org/W4402683996","https://openalex.org/W7125480319","https://openalex.org/W4391871157","https://openalex.org/W4399795004","https://openalex.org/W4411449755","https://openalex.org/W4390722860","https://openalex.org/W4403577576"],"abstract_inverted_index":{"Hallucination":[0],"in":[1,10,15,67],"large":[2],"language":[3],"models":[4],"(LLMs)":[5],"has":[6],"been":[7],"widely":[8],"studied":[9],"recent":[11],"years,":[12],"with":[13],"progress":[14],"both":[16],"detection":[17],"and":[18,44,74,120,145,149,158,166],"mitigation":[19],"aimed":[20],"at":[21,52],"improving":[22],"truthfulness.Yet,":[23],"a":[24,108],"critical":[25],"side":[26],"effect":[27],"remains":[28],"largely":[29],"overlooked:":[30],"enhancing":[31],"truthfulness":[32,165],"can":[33,97],"negatively":[34],"impact":[35],"safety":[36,136],"alignment.In":[37],"this":[38,42,62],"paper,":[39],"we":[40,106],"investigate":[41],"trade-off":[43,163],"show":[45],"that":[46,61,70,110,152],"increasing":[47,133],"factual":[48,82],"accuracy":[49],"often":[50],"comes":[51],"the":[53,68,101,162],"cost":[54],"of":[55],"weakened":[56],"refusal":[57,75,122,156],"behavior.Our":[58],"analysis":[59],"reveals":[60],"arises":[63],"from":[64,114,132],"overlapping":[65],"components":[66],"model":[69],"simultaneously":[71],"encode":[72],"hallucination":[73,115],"information,":[76],"leading":[77],"alignment":[78,99],"methods":[79],"to":[80],"suppress":[81],"knowledge":[83],"unintentionally.We":[84],"further":[85],"examine":[86],"how":[87],"fine-tuning":[88,125],"on":[89,141],"benign":[90],"datasets,":[91],"even":[92],"when":[93],"curated":[94],"for":[95,100],"safety,":[96],"degrade":[98],"same":[102],"reason.To":[103],"address":[104],"this,":[105],"propose":[107],"method":[109,140],"disentangles":[111],"refusal-related":[112],"features":[113,116],"using":[117],"sparse":[118],"autoencoders,":[119],"preserves":[121,155],"behavior":[123,157],"during":[124],"through":[126],"subspace":[127],"orthogonalization.This":[128],"approach":[129,154],"prevents":[130],"hallucinations":[131],"while":[134],"maintaining":[135],"alignment.We":[137],"evaluate":[138],"our":[139,153],"commonsense":[142],"reasoning":[143],"tasks":[144],"harmful":[146],"benchmarks":[147],"(AdvBench":[148],"StrongReject).Results":[150],"demonstrate":[151],"task":[159],"utility,":[160],"mitigating":[161],"between":[164],"safety.":[167],"1":[168]},"counts_by_year":[{"year":2026,"cited_by_count":2}],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2026-02-04T00:00:00"}