{"id":"https://openalex.org/W7167283581","doi":"https://doi.org/10.48550/arxiv.2607.01313","title":"Black-Box Inference of LLM Architectural Properties with Restrictive API Access","display_name":"Black-Box Inference of LLM Architectural Properties with Restrictive API Access","publication_year":2026,"publication_date":"2026-07-01","ids":{"openalex":"https://openalex.org/W7167283581","doi":"https://doi.org/10.48550/arxiv.2607.01313"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2607.01313","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.01313","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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.2607.01313","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5140018595","display_name":"Christopher Ellis","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ellis, Christopher","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139961109","display_name":"Shreyas Chaudhari","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chaudhari, Shreyas","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5078914856","display_name":"M Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Mei-Yu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5094008598","display_name":"Leighton Barnes","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Barnes, Leighton","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139951108","display_name":"Giulia Fanti","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fanti, Giulia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5140012305","display_name":"Jos\u00e9 M. F. Moura","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Moura, Jos\u00e9 M. F.","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/T11181","display_name":"Advanced Data Storage Technologies","score":0.12250000238418579,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.12250000238418579,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T12479","display_name":"Web Application Security Vulnerabilities","score":0.07569999992847443,"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/T12127","display_name":"Software System Performance and Reliability","score":0.06800000369548798,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/security-token","display_name":"Security token","score":0.7401000261306763},{"id":"https://openalex.org/keywords/dimension","display_name":"Dimension (graph theory)","score":0.7026000022888184},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.6377000212669373},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6161999702453613},{"id":"https://openalex.org/keywords/scale","display_name":"Scale (ratio)","score":0.4375},{"id":"https://openalex.org/keywords/data-set","display_name":"Data set","score":0.3418999910354614},{"id":"https://openalex.org/keywords/logit","display_name":"Logit","score":0.32600000500679016}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7401000261306763},{"id":"https://openalex.org/C33676613","wikidata":"https://www.wikidata.org/wiki/Q13415176","display_name":"Dimension (graph theory)","level":2,"score":0.7026000022888184},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6940000057220459},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.6377000212669373},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6161999702453613},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.4375},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.38830000162124634},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.3418999910354614},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.32749998569488525},{"id":"https://openalex.org/C140331021","wikidata":"https://www.wikidata.org/wiki/Q1868104","display_name":"Logit","level":2,"score":0.32600000500679016},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3107999861240387},{"id":"https://openalex.org/C23224414","wikidata":"https://www.wikidata.org/wiki/Q176769","display_name":"Hidden Markov model","level":2,"score":0.29499998688697815},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.29120001196861267},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.27790001034736633},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.26899999380111694},{"id":"https://openalex.org/C114289077","wikidata":"https://www.wikidata.org/wiki/Q3284399","display_name":"Statistical model","level":2,"score":0.2606000006198883},{"id":"https://openalex.org/C45804977","wikidata":"https://www.wikidata.org/wiki/Q7239673","display_name":"Predictive modelling","level":2,"score":0.2547999918460846},{"id":"https://openalex.org/C148043351","wikidata":"https://www.wikidata.org/wiki/Q4456944","display_name":"Current (fluid)","level":2,"score":0.2506999969482422},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.25040000677108765}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2607.01313","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.01313","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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.2607.01313","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.01313","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"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":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"In":[0],"practice,":[1],"most":[2,60],"commercial":[3,61],"LLM":[4,13,27,62,243],"providers":[5,63],"do":[6],"not":[7,246],"publicly":[8],"release":[9],"details":[10,41,254],"of":[11,42,50,123,147,153,255],"underlying":[12,257],"architectures.":[14],"However,":[15],"prior":[16],"work":[17],"has":[18],"shown":[19],"that":[20,91,108,241],"given":[21],"limited":[22],"API":[23,112],"access":[24,113],"to":[25,57,68,86,114,158,167,177,194,212,225,249],"an":[26,43,106,124],"(namely,":[28],"top-$k$":[29],"logits":[30],"and/or":[31],"a":[32,130,150],"logit":[33,73],"bias":[34,87],"function),":[35],"one":[36],"can":[37],"recover":[38],"certain":[39],"architectural":[40,98,253],"LLM,":[44],"such":[45],"as":[46],"the":[47,51,71,84,116,144,173,252],"hidden":[48,117,160,175,192],"dimension":[49,176,193],"feed-forward":[52],"network.":[53],"Perhaps":[54],"in":[55,136],"response":[56],"these":[58,154,228],"results,":[59],"have":[64],"restricted":[65,248],"their":[66,256],"APIs":[67,244],"expose":[69,140],"only":[70],"single":[72],"for":[74,143,215],"each":[75],"decoded":[76],"token,":[77],"and":[78,120,172,180,207,209,234],"they":[79],"no":[80],"longer":[81],"give":[82],"users":[83],"ability":[85],"logits.":[88],"We":[89,103,183,221],"show":[90],"even":[92],"under":[93],"current":[94,242],"restrictive":[95,110],"APIs,":[96],"several":[97],"parameters":[99],"are":[100,245],"still":[101],"recoverable.":[102],"present":[104],"NightVision,":[105],"attack":[107],"uses":[109,164],"black-box":[111],"estimate":[115,178],"dimension,":[118],"depth,":[119],"parameter":[121,181,210],"count":[122,211],"LLM.":[125],"Algorithmically,":[126],"NightVision":[127,162,186],"relies":[128],"on":[129,187,204],"novel":[131],"common":[132],"set":[133,146],"prompting":[134],"technique":[135],"which":[137],"multiple":[138],"prompts":[139],"log":[141],"probabilities":[142],"same":[145],"output":[148],"tokens;":[149],"spectral":[151],"analysis":[152],"results":[155,239],"is":[156],"used":[157],"infer":[159],"dimension.":[161],"additionally":[163],"end-to-end":[165],"time":[166],"first":[168],"token":[169,232],"(TTFT)":[170],"measurements":[171],"estimated":[174],"depth":[179,208],"count.":[182],"empirically":[184],"evaluate":[185],"32":[188],"open-source":[189],"LLMs,":[190],"recovering":[191],"within":[195,213],"23%":[196],"average":[197],"relative":[198],"error":[199],"across":[200],"all":[201],"models":[202,216],"(9%":[203],"MoE":[205],"models),":[206],"53%":[214],"exceeding":[217],"three":[218],"billion":[219],"parameters.":[220],"run":[222],"extensive":[223],"ablations":[224],"demonstrate":[226],"how":[227],"accuracies":[229],"scale":[230],"with":[231],"budget":[233],"model":[235],"properties.":[236],"Overall,":[237],"our":[238],"suggest":[240],"sufficiently":[247],"fully":[250],"obfuscate":[251],"models.":[258]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-07-04T00:00:00"}