{"id":"https://openalex.org/W7156196330","doi":"https://doi.org/10.1145/3774904.3792438","title":"The Asymmetric Vulnerability: Bypassing LLM Defenses via Guardrail-Model Mismatch","display_name":"The Asymmetric Vulnerability: Bypassing LLM Defenses via Guardrail-Model Mismatch","publication_year":2026,"publication_date":"2026-04-12","ids":{"openalex":"https://openalex.org/W7156196330","doi":"https://doi.org/10.1145/3774904.3792438"},"language":null,"primary_location":{"id":"doi:10.1145/3774904.3792438","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3774904.3792438","pdf_url":null,"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 ACM Web Conference 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3774904.3792438","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5108047889","display_name":"Yi-Xiang Wang","orcid":"https://orcid.org/0000-0001-5697-0717"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Junyi Wang","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0009-0004-4131-6869","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134719955","display_name":"Zhibin Zhu","orcid":"https://orcid.org/0009-0004-6289-8773"},"institutions":[{"id":"https://openalex.org/I204983213","display_name":"Harbin Institute of Technology","ror":"https://ror.org/01yqg2h08","country_code":"CN","type":"education","lineage":["https://openalex.org/I204983213"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhibin Zhu","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China"],"raw_orcid":"https://orcid.org/0009-0004-6289-8773","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China","institution_ids":["https://openalex.org/I204983213"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5043020126","display_name":"Chuanyi Liu","orcid":"https://orcid.org/0000-0002-0891-2544"},"institutions":[{"id":"https://openalex.org/I4210136793","display_name":"Peng Cheng Laboratory","ror":"https://ror.org/03qdqbt06","country_code":"CN","type":"facility","lineage":["https://openalex.org/I4210136793"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chuanyi Liu","raw_affiliation_strings":["Harbin Institute of Technology, Shenzhen, China and Pengcheng Laboratory, Shenzhen, China"],"raw_orcid":"https://orcid.org/0000-0002-0891-2544","affiliations":[{"raw_affiliation_string":"Harbin Institute of Technology, Shenzhen, China and Pengcheng Laboratory, Shenzhen, China","institution_ids":["https://openalex.org/I4210136793"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"3078","last_page":"3089"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.8468000292778015,"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":0.8468000292778015,"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/T11424","display_name":"Security and Verification in Computing","score":0.03400000184774399,"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":0.01860000006854534,"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/adversarial-system","display_name":"Adversarial system","score":0.6481999754905701},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.49459999799728394},{"id":"https://openalex.org/keywords/representation","display_name":"Representation (politics)","score":0.4124999940395355},{"id":"https://openalex.org/keywords/empirical-evidence","display_name":"Empirical evidence","score":0.3278000056743622},{"id":"https://openalex.org/keywords/character","display_name":"Character (mathematics)","score":0.2721000015735626}],"concepts":[{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.6481999754905701},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6238999962806702},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.49459999799728394},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.45660001039505005},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.4124999940395355},{"id":"https://openalex.org/C166052673","wikidata":"https://www.wikidata.org/wiki/Q83021","display_name":"Empirical evidence","level":2,"score":0.3278000056743622},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.30399999022483826},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.28049999475479126},{"id":"https://openalex.org/C2780861071","wikidata":"https://www.wikidata.org/wiki/Q1062934","display_name":"Character (mathematics)","level":2,"score":0.2721000015735626},{"id":"https://openalex.org/C2993493390","wikidata":"https://www.wikidata.org/wiki/Q5453208","display_name":"First line","level":2,"score":0.26759999990463257},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.250900000333786}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3774904.3792438","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3774904.3792438","pdf_url":null,"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 ACM Web Conference 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.1145/3774904.3792438","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3774904.3792438","pdf_url":null,"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 ACM Web Conference 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":11,"referenced_works":["https://openalex.org/W4382318449","https://openalex.org/W4396832610","https://openalex.org/W4401863964","https://openalex.org/W4403779438","https://openalex.org/W4405181600","https://openalex.org/W4409657133","https://openalex.org/W4409671304","https://openalex.org/W4411551017","https://openalex.org/W4412073525","https://openalex.org/W7133208539","https://openalex.org/W7133211372"],"related_works":[],"abstract_inverted_index":{"The":[0],"rapid":[1],"proliferation":[2],"of":[3,48],"large":[4],"language":[5],"models":[6],"(LLMs)":[7],"in":[8,174],"web-scale":[9],"applications":[10],"has":[11],"heightened":[12],"the":[13,45,67],"need":[14],"for":[15,183],"robust":[16],"security":[17,73,177],"mechanisms.":[18],"To":[19],"address":[20],"these":[21],"challenges,":[22],"various":[23],"guardrail":[24,161],"strategies":[25],"have":[26],"been":[27],"introduced,":[28],"including":[29],"input":[30],"filters,":[31],"output":[32],"moderation,":[33],"and":[34,66,97,127,138,179,191],"model":[35],"alignment.":[36,194],"Among":[37],"them,":[38],"input-side":[39,64],"guardrails":[40,65,84,141],"are":[41,85],"commonly":[42],"adopted":[43],"as":[44],"first":[46],"line":[47],"defense":[49],"against":[50],"adversarial":[51,114],"prompts.":[52],"However,":[53],"we":[54,81,109,152],"identify":[55,153],"a":[56,60,72,112,154],"critical":[57],"architectural":[58],"weakness:":[59],"representation":[61,193],"asymmetry":[62],"between":[63],"core":[68],"LLMs,":[69],"which":[70],"creates":[71],"gap":[74],"exploitable":[75],"by":[76],"adversaries.":[77],"Through":[78],"systematic":[79],"analysis,":[80],"show":[82],"that":[83,116],"fragile":[86],"when":[87],"facing":[88],"minor":[89],"character":[90,118],"perturbations,":[91],"while":[92],"LLMs":[93,137],"remain":[94,164],"semantically":[95],"resilient":[96,186],"can":[98],"still":[99],"reconstruct":[100],"malicious":[101],"intent":[102],"from":[103],"noisy":[104],"inputs.":[105],"Exploiting":[106],"this":[107],"asymmetry,":[108],"propose":[110],"RepMism,":[111],"hybrid":[113],"framework":[115],"combines":[117],"injection":[119],"with":[120],"chain-of-thought":[121],"hijacking,":[122],"coordinated":[123],"through":[124,188],"hierarchical":[125],"scheduling":[126],"safety":[128],"continuation.":[129],"Our":[130],"extensive":[131],"empirical":[132],"evaluation":[133],"across":[134],"four":[135],"commercial":[136],"five":[139],"state-of-the-art":[140],"demonstrates":[142],"RepMism's":[143],"strong":[144],"attack":[145],"capability,":[146],"achieving":[147],"high":[148],"success":[149],"rates.":[150],"Importantly,":[151],"''success":[155],"interval''":[156],"where":[157],"perturbations":[158],"effectively":[159],"bypass":[160],"detection":[162],"yet":[163],"interpretable":[165],"to":[166],"target":[167],"models.":[168],"These":[169],"findings":[170],"expose":[171],"key":[172],"flaws":[173],"current":[175],"multi-layer":[176],"architectures":[178],"offer":[180],"actionable":[181],"guidance":[182],"building":[184],"more":[185],"defenses":[187],"perturbation-aware":[189],"training":[190],"cross-layer":[192]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-21T08:15:58.654021","created_date":"2026-04-28T00:00:00"}