{"id":"https://openalex.org/W7166832605","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1118","title":"When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Models","display_name":"When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Models","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166832605","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1118"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.1118","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1118","pdf_url":"https://aclanthology.org/2026.findings-acl.1118.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 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.1118.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5037537537","display_name":"\u6bdb\u82f1\u829d","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yingzhi Mao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139800909","display_name":"Chunkang Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chunkang Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100619909","display_name":"Junxiang Wang","orcid":"https://orcid.org/0000-0002-6635-4296"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Junxiang Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139835271","display_name":"Xinyan Guan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xinyan Guan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139808219","display_name":"Boxi Cao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Boxi Cao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139799701","display_name":"Yaojie Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yaojie Lu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139813510","display_name":"Hongyu Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hongyu Lin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139718466","display_name":"Xianpei Han","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xianpei Han","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139739506","display_name":"Le Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Le Sun","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":0.82229492,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"22274","last_page":"22302"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.33649998903274536,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.33649998903274536,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.2888999879360199,"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":0.10019999742507935,"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/key","display_name":"Key (lock)","score":0.2492000013589859},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.2492000013589859},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.24889999628067017},{"id":"https://openalex.org/keywords/case-based-reasoning","display_name":"Case-based reasoning","score":0.23469999432563782},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.23109999299049377}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5519000291824341},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3880999982357025},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2492000013589859},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.2492000013589859},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.24889999628067017},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.23469999432563782},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.23109999299049377},{"id":"https://openalex.org/C159032336","wikidata":"https://www.wikidata.org/wiki/Q2488768","display_name":"Non-monotonic logic","level":2,"score":0.225600004196167},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.22450000047683716},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.21619999408721924}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.1118","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1118","pdf_url":"https://aclanthology.org/2026.findings-acl.1118.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 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.1118","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1118","pdf_url":"https://aclanthology.org/2026.findings-acl.1118.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 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/5","score":0.42006629705429077,"display_name":"Gender equality"}],"awards":[{"id":"https://openalex.org/G2823011528","display_name":null,"funder_award_id":"62306303","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166832605.pdf","grobid_xml":"https://content.openalex.org/works/W7166832605.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Content":[0],"Warning:":[1],"This":[2],"paper":[3],"contains":[4],"examples":[5],"of":[6,53,76,98],"harmful":[7,28,74],"language.Large":[8],"Reasoning":[9],"Models":[10],"(LRMs)":[11],"achieve":[12],"strong":[13],"performance":[14,148],"on":[15],"complex":[16],"multi-step":[17],"reasoning,":[18],"yet":[19],"they":[20],"still":[21],"exhibit":[22],"severe":[23],"safety":[24,102,133,145],"failures":[25,103],"such":[26],"as":[27],"content":[29],"generation.Existing":[30],"methods":[31],"often":[32],"apply":[33],"coarse-grained":[34],"constraints":[35],"over":[36],"the":[37,50,73],"entire":[38],"reasoning":[39,44,85,107,129,135,147],"trajectories,":[40],"which":[41],"can":[42],"undermine":[43],"capability":[45],"while":[46,101,127],"failing":[47],"to":[48],"address":[49],"root":[51],"causes":[52],"unsafe":[54,89],"behavior.In":[55],"this":[56,81,110],"work,":[57],"we":[58,112],"uncover":[59],"a":[60,77,91,116,141],"previously":[61],"underexplored":[62],"failure":[63],"mode":[64],"in":[65],"LRMs,":[66],"termed":[67],"Self-Jailbreak,":[68],"where":[69],"models":[70],"initially":[71],"recognize":[72],"intent":[75],"query,":[78],"but":[79],"override":[80],"judgment":[82],"during":[83],"subsequent":[84],"steps,":[86],"ultimately":[87],"generating":[88],"outputs.Such":[90],"phenomenon":[92],"reveals":[93],"that":[94,120,138],"LRMs":[95],"are":[96],"capable":[97],"recognizing":[99],"harm,":[100],"primarily":[104],"arise":[105],"from":[106],"steps.Motivated":[108],"by":[109],"finding,":[111],"propose":[113],"Chainof-Guardrail":[114],"(CoG),":[115],"trajectory-level":[117],"training":[118],"framework":[119],"mitigates":[121],"Self-Jailbreak":[122],"via":[123],"targeted,":[124],"step-level":[125],"interventions":[126],"maintaining":[128],"ability.Experiments":[130],"across":[131],"multiple":[132],"and":[134,146],"benchmarks":[136],"indicate":[137],"CoG":[139],"achieves":[140],"favorable":[142],"balance":[143],"between":[144],"compared":[149],"with":[150],"existing":[151],"approaches.":[152]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}