{"id":"https://openalex.org/W7162458415","doi":"https://doi.org/10.48550/arxiv.2605.25534","title":"StructBreak: Structural Cognitive Overload-Induced Safety Failures in MLLMs","display_name":"StructBreak: Structural Cognitive Overload-Induced Safety Failures in MLLMs","publication_year":2026,"publication_date":"2026-05-25","ids":{"openalex":"https://openalex.org/W7162458415","doi":"https://doi.org/10.48550/arxiv.2605.25534"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.25534","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.25534","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.25534","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137065121","display_name":"Yang Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Luo, Yang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136999217","display_name":"Xinran Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xinran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137044881","display_name":"Tiantian Ji","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ji, Tiantian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137083657","display_name":"Zhiyi Yin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yin, Zhiyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135950120","display_name":"Lingyun Peng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng, Lingyun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137006428","display_name":"Shuyu Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Shuyu","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":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9363999962806702,"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":0.9363999962806702,"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":0.00570000009611249,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.004999999888241291,"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/cognition","display_name":"Cognition","score":0.6474999785423279},{"id":"https://openalex.org/keywords/mechanism","display_name":"Mechanism (biology)","score":0.44029998779296875},{"id":"https://openalex.org/keywords/component","display_name":"Component (thermodynamics)","score":0.4327999949455261},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.3950999975204468},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.36250001192092896},{"id":"https://openalex.org/keywords/root-cause","display_name":"Root cause","score":0.3246999979019165},{"id":"https://openalex.org/keywords/causal-reasoning","display_name":"Causal reasoning","score":0.3095000088214874}],"concepts":[{"id":"https://openalex.org/C169900460","wikidata":"https://www.wikidata.org/wiki/Q2200417","display_name":"Cognition","level":2,"score":0.6474999785423279},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6265000104904175},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.44029998779296875},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.4327999949455261},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3970000147819519},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3950999975204468},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.36250001192092896},{"id":"https://openalex.org/C84945661","wikidata":"https://www.wikidata.org/wiki/Q7366567","display_name":"Root cause","level":2,"score":0.3246999979019165},{"id":"https://openalex.org/C115086926","wikidata":"https://www.wikidata.org/wiki/Q17004651","display_name":"Causal reasoning","level":3,"score":0.3095000088214874},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.30559998750686646},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.30059999227523804},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.29910001158714294},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.29440000653266907},{"id":"https://openalex.org/C50335755","wikidata":"https://www.wikidata.org/wiki/Q483247","display_name":"Phenomenon","level":2,"score":0.28999999165534973},{"id":"https://openalex.org/C161407221","wikidata":"https://www.wikidata.org/wiki/Q4382939","display_name":"Cognitive model","level":3,"score":0.28189998865127563},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.2705000042915344},{"id":"https://openalex.org/C16345878","wikidata":"https://www.wikidata.org/wiki/Q107472979","display_name":"Orientation (vector space)","level":2,"score":0.27000001072883606},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.26570001244544983},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.262800008058548},{"id":"https://openalex.org/C61641136","wikidata":"https://www.wikidata.org/wiki/Q1107019","display_name":"Cognitive load","level":3,"score":0.2556000053882599}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.25534","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.25534","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.25534","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.25534","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Multimodal":[0],"Large":[1],"Language":[2],"Models":[3],"(MLLMs)":[4],"excel":[5],"at":[6],"structural":[7,17,163],"reasoning":[8,34],"yet":[9],"suffer":[10],"from":[11],"a":[12,27,74,86,102,124,161],"sharp":[13],"logical":[14],"brittleness":[15],"in":[16],"consistency.":[18],"We":[19],"term":[20],"this":[21,56,82,98],"phenomenon":[22],"Structural":[23],"Cognitive":[24],"Overload":[25],"(SCO),":[26],"byproduct":[28],"of":[29,51,138,173,187],"the":[30,49,136,170,185],"contention":[31],"between":[32],"deep":[33],"and":[35,45,151],"safety":[36,167,175],"alignment.":[37],"However,":[38],"prior":[39],"work":[40],"has":[41],"predominantly":[42],"targeted":[43],"typographic":[44],"pixel-level":[46],"perturbations,":[47],"leaving":[48],"study":[50],"SCO":[52,118],"largely":[53],"unexplored.":[54],"To":[55,134],"end,":[57],"we":[58,72,96,140],"propose":[59],"StructBreak,":[60,71],"an":[61],"automated":[62],"end-to-end":[63],"framework":[64,99],"designed":[65],"to":[66,100,129,165],"quantify":[67],"SCO.":[68],"By":[69],"leveraging":[70],"uncover":[73],"novel":[75,162],"higher-order":[76],"cognitive":[77],"overload":[78],"attack":[79,83],"paradigm;":[80],"notably,":[81],"operates":[84],"under":[85],"practical":[87],"black-box":[88],"setting,":[89],"requiring":[90],"no":[91],"internal":[92],"model":[93],"access.":[94],"Consequently,":[95],"utilize":[97],"establish":[101],"comprehensive":[103],"benchmark":[104],"spanning":[105,145],"ten":[106],"diverse":[107],"threat":[108],"scenarios.":[109],"Empirical":[110],"evaluations":[111],"on":[112,131],"six":[113],"leading":[114],"MLLMs":[115],"reveal":[116,156],"that":[117,157,178],"readily":[119],"triggers":[120],"toxic":[121],"generation,":[122],"yielding":[123],"92%":[125],"average":[126],"ASR":[127],"(up":[128],"97%":[130],"Gemini":[132],"2.5).":[133],"elucidate":[135],"mechanism":[137],"SCO,":[139],"further":[141],"conduct":[142],"model-level":[143],"interpretations":[144],"attention":[146],"dynamics,":[147],"latent":[148],"space":[149],"topology,":[150],"geometric":[152],"analysis.":[153],"Our":[154],"findings":[155],"StructBreak":[158],"acts":[159],"as":[160],"channel":[164],"circumvent":[166],"filters.":[168],"Furthermore,":[169],"limited":[171],"efficacy":[172],"inherent":[174],"mechanisms":[176],"underscores":[177],"current":[179],"alignment":[180],"paradigms":[181],"are":[182],"insufficient":[183],"for":[184],"era":[186],"complex":[188],"multimodal":[189],"reasoning.":[190]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-27T00:00:00"}