{"id":"https://openalex.org/W7163680376","doi":"https://doi.org/10.48550/arxiv.2606.06286","title":"LLMs Can Leak Training Data But Do They Want To? A Propensity-Aware Evaluation of Memorization in LLMs","display_name":"LLMs Can Leak Training Data But Do They Want To? A Propensity-Aware Evaluation of Memorization in LLMs","publication_year":2026,"publication_date":"2026-06-04","ids":{"openalex":"https://openalex.org/W7163680376","doi":"https://doi.org/10.48550/arxiv.2606.06286"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.06286","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06286","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":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.06286","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137959376","display_name":"Gianluca Barmina","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Barmina, Gianluca","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137977220","display_name":"Peter Schneider-Kamp","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Schneider-Kamp, Peter","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137930598","display_name":"Lukas Galke Poech","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Poech, Lukas Galke","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/T10028","display_name":"Topic Modeling","score":0.37619999051094055,"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/T10028","display_name":"Topic Modeling","score":0.37619999051094055,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.36550000309944153,"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/T13629","display_name":"Text Readability and Simplification","score":0.05139999836683273,"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/memorization","display_name":"Memorization","score":0.9180999994277954},{"id":"https://openalex.org/keywords/audit","display_name":"Audit","score":0.6467000246047974},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.387800008058548},{"id":"https://openalex.org/keywords/design-for-manufacturability","display_name":"Design for manufacturability","score":0.38370001316070557},{"id":"https://openalex.org/keywords/pipeline","display_name":"Pipeline (software)","score":0.35370001196861267},{"id":"https://openalex.org/keywords/causality","display_name":"Causality (physics)","score":0.31119999289512634},{"id":"https://openalex.org/keywords/tracing","display_name":"Tracing","score":0.30559998750686646}],"concepts":[{"id":"https://openalex.org/C30038468","wikidata":"https://www.wikidata.org/wiki/Q4354775","display_name":"Memorization","level":2,"score":0.9180999994277954},{"id":"https://openalex.org/C199521495","wikidata":"https://www.wikidata.org/wiki/Q181487","display_name":"Audit","level":2,"score":0.6467000246047974},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.567799985408783},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.387800008058548},{"id":"https://openalex.org/C62064638","wikidata":"https://www.wikidata.org/wiki/Q553878","display_name":"Design for manufacturability","level":2,"score":0.38370001316070557},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36010000109672546},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3546999990940094},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.35370001196861267},{"id":"https://openalex.org/C64357122","wikidata":"https://www.wikidata.org/wiki/Q1149766","display_name":"Causality (physics)","level":2,"score":0.31119999289512634},{"id":"https://openalex.org/C138673069","wikidata":"https://www.wikidata.org/wiki/Q322229","display_name":"Tracing","level":2,"score":0.30559998750686646},{"id":"https://openalex.org/C31170391","wikidata":"https://www.wikidata.org/wiki/Q188619","display_name":"Hierarchy","level":2,"score":0.3034999966621399},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.28839999437332153},{"id":"https://openalex.org/C2778137410","wikidata":"https://www.wikidata.org/wiki/Q2732820","display_name":"Government (linguistics)","level":2,"score":0.2863999903202057},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.28600001335144043},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.2745000123977661},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.26980000734329224},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.26429998874664307},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.2639999985694885},{"id":"https://openalex.org/C17923572","wikidata":"https://www.wikidata.org/wiki/Q7250160","display_name":"Propensity score matching","level":2,"score":0.25999999046325684},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.25200000405311584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.06286","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06286","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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.06286","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.06286","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":"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":{"Large":[0],"language":[1],"models":[2,14,137],"can":[3,15,138,179],"reproduce":[4],"training":[5,80,140,183],"data,":[6],"but":[7,145],"existing":[8,55],"memorization":[9,37,88,122,168,170,177,195],"evaluations":[10],"mostly":[11],"measure":[12],"whether":[13,23],"be":[16],"forced":[17],"to":[18,54,58,78,208],"do":[19,25,147],"so,":[20],"rather":[21],"than":[22,124],"they":[24],"so":[26,148],"under":[27],"ordinary":[28,203],"use.":[29],"We":[30,47,62,154],"introduce":[31,64],"PropMe,":[32],"a":[33,49,66,110,210],"propensity-aware":[34],"framework":[35],"for":[36,172],"evaluation":[38],"that":[39,73,157,176,194],"contrasts":[40],"prefix-based":[41],"capability":[42,114,178],"attacks":[43,118],"with":[44],"non-adversarial":[45,152],"evaluations.":[46],"propose":[48],"metric":[50],"transformation":[51],"that,":[52],"applied":[53],"functions,":[56],"allows":[57],"create":[59],"propensity":[60,130,171,205],"metrics.":[61,89],"further":[63],"SimpleTrace,":[65],"lightweight":[67],"tracing":[68],"pipeline":[69],"built":[70],"on":[71,98],"infini-gram":[72],"deterministically":[74],"attributes":[75],"model":[76],"generations":[77],"large-scale":[79],"corpora":[81],"and":[82,86,95,103,115,169,191,202],"computes":[83],"verbatim,":[84],"near-verbatim,":[85],"propensity-transformed":[87],"Evaluating":[90],"two":[91,99,106],"fully-open":[92],"models:":[93],"Comma":[94],"DFM":[96,158],"Decoder":[97],"datasets:":[100],"Common":[101,173],"Pile":[102],"Dynaword":[104],"in":[105,149,206],"languages,":[107],"we":[108,192],"find":[109,156],"consistent":[111],"gap":[112],"between":[113],"propensity:":[116],"prefix":[117],"elicit":[119],"substantially":[120],"stronger":[121],"signals":[123],"generic":[125],"or":[126],"dataset-specific":[127],"prompts,":[128],"while":[129],"scores":[131],"remain":[132],"low":[133],"overall.":[134],"Thus,":[135],"the":[136],"reveal":[139],"data":[141],"when":[142,181],"directly":[143],"elicited,":[144],"rarely":[146],"more":[150,211],"common":[151],"settings.":[153],"also":[155],"Decoder,":[159],"which":[160],"is":[161],"continually":[162],"pre-trained":[163],"from":[164],"Comma,":[165],"exhibits":[166],"reduced":[167],"Pile,":[174],"confirming":[175],"decrease":[180],"later":[182],"emphasizes":[184],"partially":[185],"different":[186],"data.":[187],"Our":[188],"results":[189],"suggest,":[190],"encourage,":[193],"audits":[196],"should":[197],"report":[198],"both":[199],"worst-case":[200],"extractability":[201],"leakage":[204],"order":[207],"have":[209],"comprehensive":[212],"view":[213],"of":[214],"this":[215],"phenomenon.":[216]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-06T00:00:00"}