{"id":"https://openalex.org/W7204452369","doi":"https://doi.org/10.5281/zenodo.22124697","title":"Lightweight Prompt Injection Detection: From Centroid-Based Embedding Classification to a Hybrid Guardrail","display_name":"Lightweight Prompt Injection Detection: From Centroid-Based Embedding Classification to a Hybrid Guardrail","publication_year":2026,"publication_date":"2026-08-27","ids":{"openalex":"https://openalex.org/W7204452369","doi":"https://doi.org/10.5281/zenodo.22124697"},"language":"en","primary_location":{"id":"doi:10.5281/zenodo.22124697","doi":"https://doi.org/10.5281/zenodo.22124697","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.22124697","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"},"volume":null,"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.5281/zenodo.22124697","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5148423375","display_name":"Prashanna Tiwari","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Tiwari, Prashanna","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5148423375"],"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":0.9976000189781189,"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":0.0006000000284984708,"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/T12262","display_name":"Hate Speech and Cyberbullying Detection","score":0.0005000000237487257,"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/prompt-injection-detection","display_name":"prompt injection detection","score":0.9639999866485596},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.8709999918937683},{"id":"https://openalex.org/keywords/nearest-centroid-classification","display_name":"nearest centroid classification","score":0.7609999775886536},{"id":"https://openalex.org/keywords/embeddings","display_name":"embeddings","score":0.5759999752044678},{"id":"https://openalex.org/keywords/linear-probing","display_name":"linear probing","score":0.7429999709129333},{"id":"https://openalex.org/keywords/tf-idf-features","display_name":"TF-IDF features","score":0.7860000133514404},{"id":"https://openalex.org/keywords/stacking-ensemble","display_name":"stacking ensemble","score":0.8980000019073486},{"id":"https://openalex.org/keywords/generalization-gap","display_name":"generalization gap","score":0.6759999990463257},{"id":"https://openalex.org/keywords/real-time-defense","display_name":"real-time defense","score":0.43299999833106995}],"concepts":[{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.8396999835968018},{"id":"https://openalex.org/C146599234","wikidata":"https://www.wikidata.org/wiki/Q511093","display_name":"Centroid","level":2,"score":0.7950999736785889},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.7861999869346619},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6273000240325928},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.6007000207901001},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4880000054836273},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.4643000066280365},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.41819998621940613},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3864000141620636},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3562000095844269},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.2858999967575073},{"id":"https://openalex.org/C33347731","wikidata":"https://www.wikidata.org/wiki/Q285210","display_name":"Stacking","level":2,"score":0.2612000107765198},{"id":"https://openalex.org/C2780945871","wikidata":"https://www.wikidata.org/wiki/Q194274","display_name":"Backup","level":2,"score":0.25929999351501465}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.5281/zenodo.22124697","doi":"https://doi.org/10.5281/zenodo.22124697","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.22124697","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"},{"id":"doi:10.5281/zenodo.22124698","doi":"https://doi.org/10.5281/zenodo.22124698","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.22124698","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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"}],"best_oa_location":{"id":"doi:10.5281/zenodo.22124697","doi":"https://doi.org/10.5281/zenodo.22124697","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.22124697","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"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":[],"sustainable_development_goals_aurora":[{"score":0.7708306908607483,"display_name":"Reduced inequalities","id":"https://metadata.un.org/sdg/10"}],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7213395357","https://openalex.org/W7160821884","https://openalex.org/W4412889730","https://openalex.org/W7164820531","https://openalex.org/W7210974962","https://openalex.org/W4404342689","https://openalex.org/W4417095805","https://openalex.org/W7147135861","https://openalex.org/W7223416926","https://openalex.org/W7146985413"],"abstract_inverted_index":{"Prompt":[0],"injection":[1,24],"attacks":[2],"against":[3],"large":[4],"language":[5],"models":[6,46],"demand":[7],"robust":[8],"real-timedefenses.":[9],"This":[10],"paper":[11],"investigates":[12],"three":[13],"questions:":[14],"(1)":[15],"whether":[16,35],"embedding":[17,55],"represen-tations":[18],"contain":[19],"discriminative":[20,58],"information":[21,59],"for":[22,191],"prompt":[23],"detection,":[25],"(2)":[26],"whynearest-centroid":[27],"classification":[28],"fails":[29],"on":[30,47,156],"these":[31],"embeddings,":[32],"and":[33,104],"(3)":[34],"lightweightstatistical":[36],"features":[37,146],"offer":[38],"a":[39,95,105,115,161,177],"viable":[40],"alternative.":[41],"Through":[42],"controlled":[43],"experiments":[44],"acrossembedding":[45],"144,569":[48],"test":[49],"samples,":[50],"we":[51],"find":[52],"that":[53,65,130],"the":[54,76,135,141,143,181],"space":[56],"con-tains":[57],"(linear":[60],"probes":[61],"achieve":[62],"AUC":[63],"0.844\u20130.855)":[64],"nearest-centroid":[66],"scoring":[67],"cannot":[68],"access":[69],"(AUC":[70],"0.547).":[71],"The":[72,108],"critical":[73],"factor":[74],"is":[75],"number":[77],"ofcentroids:":[78],"4":[79],"category":[80],"centroids":[81,89],"overlap":[82],"(C":[83],"=":[84],"0.986)":[85],"while":[86],"2":[87],"class":[88],"achieveAUC":[90],"0.763\u20130.784.":[91],"We":[92,175],"then":[93],"evaluate":[94],"lightweight":[96],"hybrid":[97,109],"guardrail":[98],"combining":[99],"117handcrafted":[100],"features,":[101],"TF-IDF":[102,131],"similarity,":[103],"stacking":[106],"ensemble.":[107],"modelachieves":[110],"83.0%":[111,168],"in-distribution":[112],"accuracy":[113],"with":[114],"measured":[116],"single-threaded":[117],"CPU":[118],"infer-ence":[119],"latency":[120],"of":[121,180],"0.149":[122],"ms":[123],"per":[124],"prompt.":[125],"An":[126],"ablation":[127],"study":[128],"reveals":[129,160],"alone(85.0%":[132],"accuracy)":[133],"outperforms":[134],"full":[136],"hybrid,":[137],"indicating":[138],"that,":[139],"under":[140],"evaluatedsetting,":[142],"117":[144],"handcrafted":[145],"do":[147],"not":[148],"provide":[149],"complementary":[150],"predictive":[151],"valuebeyond":[152],"TF-IDF.":[153],"Cross-dataset":[154],"evaluation":[155],"four":[157],"external":[158],"benchmarks":[159],"se-vere":[162],"generalization":[163],"gap:":[164],"performance":[165],"drops":[166],"from":[167],"to":[169],"51.5%":[170],"(a":[171],"31.5":[172],"percentage-point":[173],"decline).":[174],"present":[176],"detailed":[178],"analysis":[179],"model\u2019s":[182],"strengths,":[183],"weaknesses,and":[184],"failure":[185],"modes,":[186],"providing":[187],"experimentally":[188],"grounded":[189],"guidance":[190],"designing":[192],"promptinjection":[193],"detection":[194],"systems.":[195]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-08-28T00:00:00"}