{"id":"https://openalex.org/W4412887948","doi":"https://doi.org/10.18653/v1/2025.findings-acl.960","title":"Safety is Not Only About Refusal: Reasoning-Enhanced Fine-tuning for Interpretable LLM Safety","display_name":"Safety is Not Only About Refusal: Reasoning-Enhanced Fine-tuning for Interpretable LLM Safety","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412887948","doi":"https://doi.org/10.18653/v1/2025.findings-acl.960"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.960","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.960","pdf_url":"https://aclanthology.org/2025.findings-acl.960.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.960.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5022185635","display_name":"Yuyou Zhang\uf020","orcid":"https://orcid.org/0000-0002-4699-2370"},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yuyou Zhang","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100724958","display_name":"Miao Li","orcid":"https://orcid.org/0000-0002-8626-7413"},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Miao Li","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5066352315","display_name":"William Han","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"William Han","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5074966539","display_name":"Yihang Yao","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yihang Yao","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5061106766","display_name":"Zhepeng Cen","orcid":null},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zhepeng Cen","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5037644321","display_name":"Ding Zhao","orcid":"https://orcid.org/0000-0002-9400-8446"},"institutions":[{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ding Zhao","raw_affiliation_strings":["Carnegie Mellon University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University","institution_ids":["https://openalex.org/I74973139"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I74973139"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":5,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"18727","last_page":"18746"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9909999966621399,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9909999966621399,"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/T10215","display_name":"Semantic Web and Ontologies","score":0.9248999953269958,"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":0.9100000262260437,"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/computer-science","display_name":"Computer science","score":0.5948103070259094},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.3381960391998291}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5948103070259094},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3381960391998291}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.960","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.960","pdf_url":"https://aclanthology.org/2025.findings-acl.960.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.960","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.960","pdf_url":"https://aclanthology.org/2025.findings-acl.960.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/F4320317153","display_name":"DeepMind","ror":"https://ror.org/00971b260"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412887948.pdf","grobid_xml":"https://content.openalex.org/works/W4412887948.grobid-xml"},"referenced_works_count":1,"referenced_works":["https://openalex.org/W4400480697"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"are":[4,31],"vulnerable":[5],"to":[6,26,65,82,131],"jailbreak":[7],"attacks":[8],"that":[9,62,95],"exploit":[10],"weaknesses":[11],"in":[12,67,79,141],"traditional":[13],"safety":[14,42,86,96],"alignment,":[15],"which":[16],"often":[17],"relies":[18],"on":[19],"rigid":[20],"refusal":[21],"heuristics":[22],"or":[23],"representation":[24],"engineering":[25],"block":[27],"harmful":[28,133],"outputs.While":[29],"they":[30,37],"effective":[32],"for":[33,103,123],"direct":[34],"adversarial":[35],"attacks,":[36],"fall":[38],"short":[39],"of":[40,116],"broader":[41],"challenges":[43],"requiring":[44,100],"nuanced,":[45],"context-aware":[46,139],"decision-making.To":[47],"address":[48],"this,":[49],"we":[50],"propose":[51],"ReAsoning-enhanced":[52],"Fine-TunIng":[53],"fOr":[54],"iNterpretAble":[55],"LLM":[56,124],"Safety":[57],"(RATIONAL),":[58],"a":[59,113,120],"novel":[60],"framework":[61],"trains":[63],"models":[64,73],"engage":[66],"explicit":[68],"safe":[69],"reasoning":[70,81],"before":[71],"response.Fine-tuned":[72],"leverage":[74],"the":[75],"extensive":[76],"pretraining":[77],"knowledge":[78],"selfgenerated":[80],"bootstrap":[83],"their":[84],"own":[85],"through":[87],"structured":[88],"reasoning,":[89],"internalizing":[90],"context-sensitive":[91],"decision-making.Our":[92],"findings":[93],"suggest":[94],"extends":[97],"beyond":[98],"refusal,":[99],"context":[101],"awareness":[102],"more":[104],"robust,":[105],"interpretable,":[106],"and":[107,138],"adaptive":[108],"responses.Reasoning":[109],"is":[110],"not":[111],"only":[112],"core":[114],"capability":[115],"LLMs":[117],"but":[118],"also":[119],"fundamental":[121],"mechanism":[122],"safety.RATIONAL":[125],"employs":[126],"reasoningenhanced":[127],"fine-tuning,":[128],"allowing":[129],"it":[130],"reject":[132],"prompts":[134],"while":[135],"providing":[136],"meaningful":[137],"responses":[140],"complex":[142],"scenarios.":[143]},"counts_by_year":[{"year":2026,"cited_by_count":2},{"year":2025,"cited_by_count":3}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}