{"id":"https://openalex.org/W4416035657","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.1392","title":"MrGuard: A Multilingual Reasoning Guardrail for Universal LLM Safety","display_name":"MrGuard: A Multilingual Reasoning Guardrail for Universal LLM Safety","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416035657","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.1392"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.emnlp-main.1392","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1392","pdf_url":"https://aclanthology.org/2025.emnlp-main.1392.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.emnlp-main.1392.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120308875","display_name":"Yahan Yang","orcid":null},"institutions":[{"id":"https://openalex.org/I36788626","display_name":"California University of Pennsylvania","ror":"https://ror.org/01spssf70","country_code":"US","type":"education","lineage":["https://openalex.org/I36788626"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yahan Yang","raw_affiliation_strings":["University of Pennsylvania"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Pennsylvania","institution_ids":["https://openalex.org/I36788626"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120308876","display_name":"Soham Dan","orcid":null},"institutions":[{"id":"https://openalex.org/I4210164937","display_name":"Microsoft Research (United Kingdom)","ror":"https://ror.org/05k87vq12","country_code":"GB","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210164937"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Soham Dan","raw_affiliation_strings":["Microsoft"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft","institution_ids":["https://openalex.org/I4210164937"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120308877","display_name":"Shuo Li","orcid":null},"institutions":[{"id":"https://openalex.org/I36788626","display_name":"California University of Pennsylvania","ror":"https://ror.org/01spssf70","country_code":"US","type":"education","lineage":["https://openalex.org/I36788626"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Shuo Li","raw_affiliation_strings":["University of Pennsylvania"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Pennsylvania","institution_ids":["https://openalex.org/I36788626"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120308878","display_name":"Dan Roth","orcid":null},"institutions":[{"id":"https://openalex.org/I1342911587","display_name":"Oracle (United States)","ror":"https://ror.org/006c77m33","country_code":"US","type":"company","lineage":["https://openalex.org/I1342911587"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Dan Roth","raw_affiliation_strings":["University of Pennsylvania Oracle AI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Pennsylvania Oracle AI","institution_ids":["https://openalex.org/I1342911587"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5120308879","display_name":"Insup Lee","orcid":null},"institutions":[{"id":"https://openalex.org/I36788626","display_name":"California University of Pennsylvania","ror":"https://ror.org/01spssf70","country_code":"US","type":"education","lineage":["https://openalex.org/I36788626"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Insup Lee","raw_affiliation_strings":["University of Pennsylvania"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Pennsylvania","institution_ids":["https://openalex.org/I36788626"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.4041,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.84611474,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"27365","last_page":"27384"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.1476999968290329,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.1476999968290329,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.1185000017285347,"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":0.07980000227689743,"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/action","display_name":"Action (physics)","score":0.3280999958515167},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.2565999925136566},{"id":"https://openalex.org/keywords/feature","display_name":"Feature (linguistics)","score":0.2551000118255615},{"id":"https://openalex.org/keywords/natural-language","display_name":"Natural language","score":0.2531999945640564}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5223000049591064},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.3472000062465668},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.3280999958515167},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3246999979019165},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3131999969482422},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.27630001306533813},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2565999925136566},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.2551000118255615},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.2531999945640564},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.23499999940395355}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.emnlp-main.1392","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1392","pdf_url":"https://aclanthology.org/2025.emnlp-main.1392.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.emnlp-main.1392","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1392","pdf_url":"https://aclanthology.org/2025.emnlp-main.1392.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G4785748979","display_name":null,"funder_award_id":"W911NF2010080","funder_id":"https://openalex.org/F4320338281","funder_display_name":"Army Research Office"},{"id":"https://openalex.org/G7452299184","display_name":null,"funder_award_id":"W911NF","funder_id":"https://openalex.org/F4320338281","funder_display_name":"Army Research Office"}],"funders":[{"id":"https://openalex.org/F4320338281","display_name":"Army Research Office","ror":"https://ror.org/05epdh915"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416035657.pdf","grobid_xml":"https://content.openalex.org/works/W4416035657.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"are":[4,157],"susceptible":[5],"to":[6,121,153],"adversarial":[7],"attacks":[8],"such":[9,124],"as":[10,125],"jailbreaking,":[11],"which":[12,156],"can":[13],"elicit":[14],"harmful":[15],"or":[16],"unsafe":[17,40],"behaviors.This":[18],"vulnerability":[19],"is":[20,29,45],"exacerbated":[21],"in":[22,50,131,166],"multilingual":[23,26,58,70,100,122,145,167],"settings,":[24],"where":[25],"safetyaligned":[27],"data":[28,71],"often":[30],"limited.Thus,":[31],"developing":[32],"a":[33,57,84],"guardrail":[34,59,150],"capable":[35],"of":[36,148],"detecting":[37],"and":[38,75,82,110,127,134,164],"filtering":[39],"content":[41,168],"across":[42,107],"diverse":[43],"languages":[44,112],"critical":[46],"for":[47,62,160],"deploying":[48],"LLMs":[49],"real-world":[51],"applications.In":[52],"this":[53],"work,":[54],"we":[55],"introduce":[56],"with":[60],"reasoning":[61,146],"prompt":[63],"classification.Our":[64],"method":[65],"consists":[66],"of:":[67],"(1)":[68],"synthetic":[69],"generation":[72],"incorporating":[73],"culturally":[74],"linguistically":[76],"nuanced":[77],"variants,":[78],"(2)":[79],"supervised":[80],"fine-tuning,":[81],"(3)":[83],"curriculum-based":[85],"Group":[86],"Relative":[87],"Policy":[88],"Optimization":[89],"(GRPO)":[90],"framework":[91],"that":[92,98,136],"further":[93],"improves":[94],"performance.Experimental":[95],"results":[96],"demonstrate":[97,135],"our":[99,149],"guardrail,":[101],"Mr-Guard,":[102],"consistently":[103],"outperforms":[104],"recent":[105],"baselines":[106],"both":[108],"in-domain":[109],"out-of-domain":[111],"by":[113],"more":[114],"than":[115],"15%.We":[116],"also":[117],"evaluate":[118],"MrGuard's":[119],"robustness":[120],"variations,":[123],"code-switching":[126],"low-resource":[128],"language":[129],"distractors":[130],"the":[132],"prompt,":[133],"it":[137,152],"preserves":[138],"safety":[139],"judgments":[140],"under":[141],"these":[142],"challenging":[143],"conditions.The":[144],"capability":[147],"enables":[151],"generate":[154],"explanations,":[155],"particularly":[158],"useful":[159],"understanding":[161],"languagespecific":[162],"risks":[163],"ambiguities":[165],"moderation.":[169]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-08-21T09:56:20.448147","created_date":"2025-11-08T00:00:00"}