{"id":"https://openalex.org/W7148474329","doi":"https://doi.org/10.48550/arxiv.2604.00209","title":"Do LLMs Know What Is Private Internally? Probing and Steering Contextual Privacy Norms in Large Language Model Representations","display_name":"Do LLMs Know What Is Private Internally? Probing and Steering Contextual Privacy Norms in Large Language Model Representations","publication_year":2026,"publication_date":"2026-03-31","ids":{"openalex":"https://openalex.org/W7148474329","doi":"https://doi.org/10.48550/arxiv.2604.00209"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.00209","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00209","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.00209","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5132813796","display_name":"Haoran Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Wang, Haoran","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132794428","display_name":"Li Xiong","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiong, Li","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5132819288","display_name":"Kai Shu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shu, Kai","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":3,"corresponding_author_ids":["https://openalex.org/A5132813796"],"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/T13910","display_name":"Computational and Text Analysis Methods","score":0.13230000436306,"subfield":{"id":"https://openalex.org/subfields/3300","display_name":"General Social Sciences"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T13910","display_name":"Computational and Text Analysis Methods","score":0.13230000436306,"subfield":{"id":"https://openalex.org/subfields/3300","display_name":"General Social Sciences"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.10419999808073044,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.1039000004529953,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/representation","display_name":"Representation (politics)","score":0.6628999710083008},{"id":"https://openalex.org/keywords/private-information-retrieval","display_name":"Private information retrieval","score":0.5806999802589417},{"id":"https://openalex.org/keywords/bridge","display_name":"Bridge (graph theory)","score":0.47859999537467957},{"id":"https://openalex.org/keywords/contextual-design","display_name":"Contextual design","score":0.4571000039577484},{"id":"https://openalex.org/keywords/encode","display_name":"ENCODE","score":0.4246000051498413},{"id":"https://openalex.org/keywords/context-analysis","display_name":"Context analysis","score":0.40299999713897705},{"id":"https://openalex.org/keywords/information-privacy","display_name":"Information privacy","score":0.4018000066280365},{"id":"https://openalex.org/keywords/class","display_name":"Class (philosophy)","score":0.4000999927520752}],"concepts":[{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.6628999710083008},{"id":"https://openalex.org/C108827166","wikidata":"https://www.wikidata.org/wiki/Q175975","display_name":"Internet privacy","level":1,"score":0.5821999907493591},{"id":"https://openalex.org/C99221444","wikidata":"https://www.wikidata.org/wiki/Q1532069","display_name":"Private information retrieval","level":2,"score":0.5806999802589417},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5389000177383423},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.47859999537467957},{"id":"https://openalex.org/C71611378","wikidata":"https://www.wikidata.org/wiki/Q5165191","display_name":"Contextual design","level":3,"score":0.4571000039577484},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.4246000051498413},{"id":"https://openalex.org/C52085439","wikidata":"https://www.wikidata.org/wiki/Q5165173","display_name":"Context analysis","level":3,"score":0.40299999713897705},{"id":"https://openalex.org/C123201435","wikidata":"https://www.wikidata.org/wiki/Q456632","display_name":"Information privacy","level":2,"score":0.4018000066280365},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.4000999927520752},{"id":"https://openalex.org/C102938260","wikidata":"https://www.wikidata.org/wiki/Q1999831","display_name":"Privacy policy","level":3,"score":0.39969998598098755},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.3765999972820282},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.3522999882698059},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.34279999136924744},{"id":"https://openalex.org/C169093310","wikidata":"https://www.wikidata.org/wiki/Q3702971","display_name":"Personally identifiable information","level":2,"score":0.3228999972343445},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.3165999948978424},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.3116999864578247},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.3073999881744385},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.30550000071525574},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.29190000891685486},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2903999984264374},{"id":"https://openalex.org/C18414002","wikidata":"https://www.wikidata.org/wiki/Q8354932","display_name":"Right to privacy","level":2,"score":0.2858000099658966},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2743000090122223},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.27300000190734863},{"id":"https://openalex.org/C137822555","wikidata":"https://www.wikidata.org/wiki/Q2587068","display_name":"Information sensitivity","level":2,"score":0.26460000872612},{"id":"https://openalex.org/C67497173","wikidata":"https://www.wikidata.org/wiki/Q977345","display_name":"Legal aspects of computing","level":3,"score":0.25060001015663147}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.00209","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00209","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.00209","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.00209","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article"},"sustainable_development_goals":[{"score":0.7800788283348083,"display_name":"Peace, Justice and strong institutions","id":"https://metadata.un.org/sdg/16"}],"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":[2,101],"(LLMs)":[3],"are":[4,85],"increasingly":[5],"deployed":[6],"in":[7,20,60,63,94,106,187],"high-stakes":[8],"settings,":[9],"yet":[10],"they":[11],"frequently":[12],"violate":[13],"contextual":[14,36,53,64,150,175,184],"privacy":[15,37,54,137,151,176,185],"by":[16],"disclosing":[17],"private":[18,104],"information":[19,105],"situations":[21],"where":[22],"humans":[23],"would":[24],"exercise":[25],"discretion.":[26],"This":[27,133],"raises":[28],"a":[29,56,109],"fundamental":[30],"question:":[31],"do":[32,43],"LLMs":[33],"internally":[34],"encode":[35],"norms,":[38],"and":[39,82,90,115,141,158,164],"if":[40],"so,":[41],"why":[42],"violations":[44,138],"persist?":[45],"We":[46],"present":[47],"the":[48,74,167],"first":[49],"systematic":[50],"study":[51],"of":[52,170,183],"as":[55,87],"structured":[57,134],"latent":[58],"representation":[59,114,157],"LLMs,":[61],"grounded":[62],"integrity":[65],"(CI)":[66],"theory.":[67],"Probing":[68],"multiple":[69],"models,":[70],"we":[71,122],"find":[72],"that":[73,149,165],"three":[75],"norm-determining":[76],"CI":[77,131,171],"parameters":[78],"(information":[79],"type,":[80],"recipient,":[81],"transmission":[83],"principle)":[84],"encoded":[86],"linearly":[88],"separable":[89],"functionally":[91],"independent":[92],"directions":[93],"activation":[95],"space.":[96],"Despite":[97],"this":[98,120],"internal":[99],"structure,":[100],"still":[102],"leak":[103],"practice,":[107],"revealing":[108],"clear":[110],"gap":[111],"between":[112,156],"concept":[113],"model":[116],"behavior.":[117],"To":[118],"bridge":[119],"gap,":[121],"introduce":[123],"CI-parametric":[124],"steering,":[125],"which":[126],"independently":[127],"intervenes":[128],"along":[129],"each":[130],"dimension.":[132],"control":[135],"reduces":[136],"more":[139,173],"effectively":[140],"predictably":[142],"than":[143,161],"monolithic":[144],"steering.":[145],"Our":[146],"results":[147],"demonstrate":[148],"failures":[152],"arise":[153],"from":[154],"misalignment":[155],"behavior":[159],"rather":[160],"missing":[162],"awareness,":[163],"leveraging":[166],"compositional":[168],"structure":[169],"enables":[172],"reliable":[174],"control,":[177],"shedding":[178],"light":[179],"on":[180],"potential":[181],"improvement":[182],"understanding":[186],"LLMs.":[188]},"counts_by_year":[],"updated_date":"2026-04-03T16:44:17.987007","created_date":"2026-04-03T00:00:00"}