{"id":"https://openalex.org/W7202216485","doi":"https://doi.org/10.32388/7hxkqe","title":"In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning","display_name":"In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document Poisoning","publication_year":2026,"publication_date":"2026-08-11","ids":{"openalex":"https://openalex.org/W7202216485","doi":"https://doi.org/10.32388/7hxkqe"},"language":"en","primary_location":{"id":"doi:10.32388/7hxkqe","doi":"https://doi.org/10.32388/7hxkqe","is_oa":true,"landing_page_url":"https://doi.org/10.32388/7hxkqe","pdf_url":"https://www.qeios.com/read/7HXKQE/pdf","source":{"id":"https://openalex.org/S4210177074","display_name":"Qeios","issn_l":"2632-3834","issn":["2632-3834"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":["anvur-scientific","norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Qeios","raw_type":"posted-content"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://www.qeios.com/read/7HXKQE/pdf","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5140414009","display_name":"Iliano Fasolino","orcid":"https://orcid.org/0009-0009-4807-3093","observed_orcids":["https://orcid.org/0009-0009-4807-3093"]},"institutions":[{"id":"https://openalex.org/I189158943","display_name":"University of Milan","ror":"https://ror.org/00wjc7c48","country_code":"IT","type":"education","lineage":["https://openalex.org/I189158943"]}],"countries":["IT"],"is_corresponding":true,"raw_author_name":"Iliano Fasolino","raw_affiliation_strings":["University of Milan"],"raw_orcid":"https://orcid.org/0009-0009-4807-3093","affiliations":[{"raw_affiliation_string":"University of Milan","institution_ids":["https://openalex.org/I189158943"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5140414009"],"corresponding_institution_ids":["https://openalex.org/I189158943"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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":0.9976000189781189,"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":0.9976000189781189,"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":0.0017999999690800905,"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/T11147","display_name":"Misinformation and Its Impacts","score":9.999999747378752e-05,"subfield":{"id":"https://openalex.org/subfields/3312","display_name":"Sociology and Political Science"},"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/retrieval-augmented-generation","display_name":"retrieval-augmented generation","score":1.0},{"id":"https://openalex.org/keywords/robustness","display_name":"robustness","score":0.7110000252723694},{"id":"https://openalex.org/keywords/hallucinations","display_name":"hallucinations","score":0.9649999737739563},{"id":"https://openalex.org/keywords/fact-checking","display_name":"fact-checking","score":0.6819999814033508},{"id":"https://openalex.org/keywords/negation","display_name":"negation","score":0.890999972820282},{"id":"https://openalex.org/keywords/llama-3-1-8b","display_name":"Llama-3.1-8B","score":0.8149999976158142},{"id":"https://openalex.org/keywords/quantized-language-models","display_name":"quantized language models","score":0.7310000061988831},{"id":"https://openalex.org/keywords/fever","display_name":"fever","score":0.7319999933242798}],"concepts":[{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.7106000185012817},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.650600016117096},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4982999861240387},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.4945000112056732},{"id":"https://openalex.org/C99821215","wikidata":"https://www.wikidata.org/wiki/Q1136583","display_name":"Swap (finance)","level":2,"score":0.4571000039577484},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.398499995470047},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.33160001039505005},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.3131999969482422},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.3093999922275543},{"id":"https://openalex.org/C2985684807","wikidata":"https://www.wikidata.org/wiki/Q1513879","display_name":"Text generation","level":2,"score":0.303600013256073},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.30090001225471497},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.30079999566078186},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.2994000017642975},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2581000030040741},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.2542000114917755},{"id":"https://openalex.org/C2780719617","wikidata":"https://www.wikidata.org/wiki/Q1030752","display_name":"Salient","level":2,"score":0.2540000081062317}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.32388/7hxkqe","doi":"https://doi.org/10.32388/7hxkqe","is_oa":true,"landing_page_url":"https://doi.org/10.32388/7hxkqe","pdf_url":"https://www.qeios.com/read/7HXKQE/pdf","source":{"id":"https://openalex.org/S4210177074","display_name":"Qeios","issn_l":"2632-3834","issn":["2632-3834"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":["anvur-scientific","norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Qeios","raw_type":"posted-content"}],"best_oa_location":{"id":"doi:10.32388/7hxkqe","doi":"https://doi.org/10.32388/7hxkqe","is_oa":true,"landing_page_url":"https://doi.org/10.32388/7hxkqe","pdf_url":"https://www.qeios.com/read/7HXKQE/pdf","source":{"id":"https://openalex.org/S4210177074","display_name":"Qeios","issn_l":"2632-3834","issn":["2632-3834"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":["anvur-scientific","norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Qeios","raw_type":"posted-content"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7202216485.pdf","grobid_xml":"https://content.openalex.org/works/W7202216485.grobid-xml"},"referenced_works_count":4,"referenced_works":["https://openalex.org/W2593864460","https://openalex.org/W2789566302","https://openalex.org/W2970641574","https://openalex.org/W4388444775"],"related_works":["https://openalex.org/W7212325956","https://openalex.org/W7212316521","https://openalex.org/W4405766378","https://openalex.org/W7171865459","https://openalex.org/W7166830555","https://openalex.org/W7172499403","https://openalex.org/W7171546727","https://openalex.org/W4407062832","https://openalex.org/W7162790142","https://openalex.org/W7117613103"],"abstract_inverted_index":{"Retrieval-augmented":[0],"generation":[1],"(RAG)":[2],"grounds":[3],"a":[4,15,35,44,77,84,124,155],"language":[5],"model":[6,26,134],"in":[7,111],"retrieved":[8,20,48,74],"documents,":[9],"which":[10],"reduces":[11],"hallucination":[12,147],"but":[13],"creates":[14],"new":[16,137],"attack":[17],"surface:":[18],"if":[19],"text":[21],"is":[22,50,107,143],"tampered":[23],"with,":[24],"the":[25,29,72,108,131,166],"may":[27],"repeat":[28],"falsehood.":[30],"We":[31],"study":[32],"how":[33],"much":[34],"small":[36],"quantized":[37],"model,":[38],"Llama":[39],"3.1":[40],"8B,":[41],"degrades":[42],"when":[43,99],"fraction":[45],"of":[46,71,80,113,158],"its":[47,140],"context":[49,96],"poisoned.":[51],"Three":[52],"corruption":[53,122],"strategies":[54],"are":[55,103],"tested,":[56],"entity":[57],"swap,":[58,60],"number":[59],"and":[61,120,146,160],"negation,":[62],"each":[63],"applied":[64],"to":[65,97,118,144,169],"zero,":[66],"one,":[67],"two,":[68],"or":[69],"three":[70,73,101],"passages,":[75],"over":[76],"factorial":[78],"sweep":[79],"588":[81],"runs":[82],"on":[83,94],"fact-checking":[85],"task":[86],"built":[87],"from":[88,92,116],"FEVER.":[89],"Accuracy":[90],"falls":[91],"77.9%":[93],"clean":[95],"43.5%":[98],"all":[100],"passages":[102,129],"corrupted.":[104],"Entity":[105],"swap":[106],"most":[109],"damaging":[110],"terms":[112],"answers":[114],"flipped":[115],"correct":[117],"wrong,":[119],"number-based":[121],"shows":[123],"sharp":[125],"threshold":[126],"once":[127],"poisoned":[128],"form":[130],"majority.":[132],"The":[133],"rarely":[135],"invents":[136],"falsehoods;":[138],"instead":[139],"dominant":[141],"reaction":[142],"abstain,":[145],"actually":[148],"drops":[149],"under":[150],"attack.":[151],"These":[152],"results":[153],"quantify":[154],"practical":[156],"weakness":[157],"RAG":[159],"identify":[161],"abstention,":[162],"not":[163],"fabrication,":[164],"as":[165],"main":[167],"behaviour":[168],"plan":[170],"for.":[171]},"counts_by_year":[],"updated_date":"2026-11-04T10:54:05.112534","created_date":"2026-08-12T00:00:00"}