{"id":"https://openalex.org/W7162797852","doi":"https://doi.org/10.48550/arxiv.2605.29901","title":"Dissecting the Black Box: Circuit-Level Analysis of LLM Vulnerability Detection","display_name":"Dissecting the Black Box: Circuit-Level Analysis of LLM Vulnerability Detection","publication_year":2026,"publication_date":"2026-05-28","ids":{"openalex":"https://openalex.org/W7162797852","doi":"https://doi.org/10.48550/arxiv.2605.29901"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.29901","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.29901","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.2605.29901","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5008594126","display_name":"Syafiq Al Atiiq","orcid":"https://orcid.org/0000-0002-8771-7893"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Atiiq, Syafiq Al","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137323815","display_name":"Chun Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Chun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5044464349","display_name":"Christian Gehrmann","orcid":"https://orcid.org/0000-0001-8003-200X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gehrmann, Christian","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":0.4449999928474426,"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":0.4449999928474426,"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/T10260","display_name":"Software Engineering Research","score":0.1589999943971634,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T11241","display_name":"Advanced Malware Detection Techniques","score":0.08739999681711197,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/artificial-neural-network","display_name":"Artificial neural network","score":0.5641999840736389},{"id":"https://openalex.org/keywords/vulnerability","display_name":"Vulnerability (computing)","score":0.5199000239372253},{"id":"https://openalex.org/keywords/perceptron","display_name":"Perceptron","score":0.5105999708175659},{"id":"https://openalex.org/keywords/encode","display_name":"ENCODE","score":0.5022000074386597},{"id":"https://openalex.org/keywords/deep-learning","display_name":"Deep learning","score":0.43479999899864197},{"id":"https://openalex.org/keywords/layer","display_name":"Layer (electronics)","score":0.4262000024318695},{"id":"https://openalex.org/keywords/software","display_name":"Software","score":0.4205999970436096},{"id":"https://openalex.org/keywords/trace","display_name":"TRACE (psycholinguistics)","score":0.40540000796318054},{"id":"https://openalex.org/keywords/focus","display_name":"Focus (optics)","score":0.391400009393692}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6862000226974487},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.5641999840736389},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.532800018787384},{"id":"https://openalex.org/C95713431","wikidata":"https://www.wikidata.org/wiki/Q631425","display_name":"Vulnerability (computing)","level":2,"score":0.5199000239372253},{"id":"https://openalex.org/C60908668","wikidata":"https://www.wikidata.org/wiki/Q690207","display_name":"Perceptron","level":3,"score":0.5105999708175659},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.5022000074386597},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.43479999899864197},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.4262000024318695},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.4205999970436096},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4075999855995178},{"id":"https://openalex.org/C75291252","wikidata":"https://www.wikidata.org/wiki/Q1315756","display_name":"TRACE (psycholinguistics)","level":2,"score":0.40540000796318054},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.391400009393692},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.37630000710487366},{"id":"https://openalex.org/C179518139","wikidata":"https://www.wikidata.org/wiki/Q5140297","display_name":"Coding (social sciences)","level":2,"score":0.37560001015663147},{"id":"https://openalex.org/C77637269","wikidata":"https://www.wikidata.org/wiki/Q7002051","display_name":"Neural coding","level":2,"score":0.3652999997138977},{"id":"https://openalex.org/C2779662365","wikidata":"https://www.wikidata.org/wiki/Q5416694","display_name":"Event (particle physics)","level":2,"score":0.3594000041484833},{"id":"https://openalex.org/C94915269","wikidata":"https://www.wikidata.org/wiki/Q1834857","display_name":"Detector","level":2,"score":0.3285999894142151},{"id":"https://openalex.org/C118403218","wikidata":"https://www.wikidata.org/wiki/Q43283","display_name":"Biological neural network","level":2,"score":0.31709998846054077},{"id":"https://openalex.org/C167063184","wikidata":"https://www.wikidata.org/wiki/Q1400839","display_name":"Vulnerability assessment","level":3,"score":0.3082999885082245},{"id":"https://openalex.org/C32946077","wikidata":"https://www.wikidata.org/wiki/Q618079","display_name":"Network analysis","level":2,"score":0.2903999984264374},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2833999991416931},{"id":"https://openalex.org/C66024118","wikidata":"https://www.wikidata.org/wiki/Q1122506","display_name":"Computational model","level":2,"score":0.2745000123977661},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.27379998564720154},{"id":"https://openalex.org/C179717631","wikidata":"https://www.wikidata.org/wiki/Q2991667","display_name":"Multilayer perceptron","level":3,"score":0.27309998869895935},{"id":"https://openalex.org/C2776151529","wikidata":"https://www.wikidata.org/wiki/Q3045304","display_name":"Object detection","level":3,"score":0.2671000063419342},{"id":"https://openalex.org/C205427263","wikidata":"https://www.wikidata.org/wiki/Q1931480","display_name":"Neuroinformatics","level":2,"score":0.2646999955177307},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2632000148296356},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.2597000002861023},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.2554999887943268}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.29901","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.29901","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.2605.29901","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.29901","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":[{"id":"https://metadata.un.org/sdg/10","score":0.5097119808197021,"display_name":"Reduced inequalities"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"can":[4],"detect":[5],"software":[6],"vulnerabilities,":[7],"but":[8],"how":[9],"do":[10],"they":[11],"actually":[12],"identify":[13,98],"vulnerable":[14,55],"code?":[15],"We":[16,97],"address":[17],"this":[18],"question":[19],"using":[20],"mechanistic":[21],"interpretability;":[22],"analyzing":[23],"the":[24,42,47,64,91,99],"internal":[25],"computations":[26],"of":[27,169],"a":[28,61],"neural":[29,101],"network":[30],"to":[31,89,143,181],"understand":[32],"its":[33],"reasoning":[34],"process.Using":[35],"Circuit":[36],"Tracer":[37],"on":[38,68,113],"Gemma-2-2b,":[39],"we":[40],"trace":[41],"computational":[43],"pathways":[44],"activated":[45],"when":[46],"model":[48,65,92,170],"classifies":[49,93],"472":[50],"C/C++":[51],"code":[52,94],"samples":[53],"as":[54,95],"or":[56],"safe.":[57],"Our":[58],"analysis":[59],"reveals":[60],"surprising":[62],"finding:":[63],"primarily":[66],"relies":[67],"safety":[69,86,114],"detectors,":[70],"attention":[71,104],"heads":[72,105],"that":[73,111,124,159],"recognize":[74],"safe":[75],"coding":[76],"patterns,":[77,115],"rather":[78],"than":[79],"directly":[80],"detecting":[81],"vulnerability":[82,138,161],"signatures.":[83],"When":[84],"these":[85],"detectors":[87],"fail":[88],"activate,":[90],"vulnerable.":[96],"critical":[100],"components:":[102],"specific":[103],"in":[106,121,150],"early":[107],"layers":[108],"(L5,":[109],"L7)":[110],"focus":[112],"and":[116,178],"Multilayer":[117],"Perceptron":[118],"(MLP)":[119],"neurons":[120,149],"Layer":[122,135,151],"7":[123,152],"encode":[125],"vulnerability-related":[126],"features.":[127],"Ablation":[128],"experiments":[129],"confirm":[130],"their":[131],"causal":[132],"role;":[133],"removing":[134],"11":[136],"drops":[137],"detection":[139,162,182],"accuracy":[140],"from":[141],"100%":[142],"6%,":[144],"while":[145],"ablating":[146],"just":[147],"20":[148],"reduces":[153],"it":[154],"by":[155],"50%.Our":[156],"findings":[157],"show":[158],"LLM":[160],"uses":[163],"sparse,":[164],"interpretable":[165],"circuits":[166],"(only":[167],"16%":[168],"capacity),":[171],"enabling":[172],"circuit-level":[173],"explanations":[174],"for":[175],"security":[176],"predictions":[177],"targeted":[179],"improvements":[180],"systems.":[183]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-30T00:00:00"}