{"id":"https://openalex.org/W4404342689","doi":"https://doi.org/10.48550/arxiv.2410.22770","title":"InjecGuard: Benchmarking and Mitigating Over-defense in Prompt Injection Guardrail Models","display_name":"InjecGuard: Benchmarking and Mitigating Over-defense in Prompt Injection Guardrail Models","publication_year":2024,"publication_date":"2024-10-30","ids":{"openalex":"https://openalex.org/W4404342689","doi":"https://doi.org/10.48550/arxiv.2410.22770"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2410.22770","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.22770","pdf_url":"https://arxiv.org/pdf/2410.22770","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2410.22770","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101458777","display_name":"Hao Li","orcid":"https://orcid.org/0000-0003-0092-5145","observed_orcids":["https://orcid.org/0000-0003-0092-5145"]},"institutions":[{"id":"https://openalex.org/I204465549","display_name":"Washington University in St. Louis","ror":"https://ror.org/01yc7t268","country_code":"US","type":"education","lineage":["https://openalex.org/I204465549"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Li, Hao","raw_affiliation_strings":["Washington University in St. Louis ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Washington University in St. Louis ,","institution_ids":["https://openalex.org/I204465549"]}]},{"author_position":"last","author":{"id":null,"display_name":"Liu, Xiaogeng","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I135310074","display_name":"University of Wisconsin\u2013Madison","ror":"https://ror.org/01y2jtd41","country_code":"US","type":"education","lineage":["https://openalex.org/I135310074"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Liu, Xiaogeng","raw_affiliation_strings":["University of Wisconsin-Madison"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Wisconsin-Madison","institution_ids":["https://openalex.org/I135310074"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":56,"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":0.9994999766349792,"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":0.9994999766349792,"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":9.999999747378752e-05,"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":9.999999747378752e-05,"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","display_name":"prompt injection","score":0.9300000071525574},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.6840000152587891},{"id":"https://openalex.org/keywords/attack-detection","display_name":"attack detection","score":0.6890000104904175},{"id":"https://openalex.org/keywords/data-leakage","display_name":"data leakage","score":0.9110000133514404},{"id":"https://openalex.org/keywords/training-strategies","display_name":"training strategies","score":0.4020000100135803}],"concepts":[{"id":"https://openalex.org/C86251818","wikidata":"https://www.wikidata.org/wiki/Q816754","display_name":"Benchmarking","level":2,"score":0.8412630558013916},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3212316632270813},{"id":"https://openalex.org/C144133560","wikidata":"https://www.wikidata.org/wiki/Q4830453","display_name":"Business","level":0,"score":0.28840115666389465},{"id":"https://openalex.org/C162853370","wikidata":"https://www.wikidata.org/wiki/Q39809","display_name":"Marketing","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2410.22770","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.22770","pdf_url":"https://arxiv.org/pdf/2410.22770","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2410.22770","doi":"https://doi.org/10.48550/arxiv.2410.22770","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2410.22770","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2410.22770","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.22770","pdf_url":"https://arxiv.org/pdf/2410.22770","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4404342689.pdf","grobid_xml":"https://content.openalex.org/works/W4404342689.grobid-xml"},"referenced_works_count":29,"referenced_works":["https://openalex.org/W2951576127","https://openalex.org/W2952984539","https://openalex.org/W3030163527","https://openalex.org/W3198192082","https://openalex.org/W3207710400","https://openalex.org/W4225468040","https://openalex.org/W4294962858","https://openalex.org/W4309395891","https://openalex.org/W4384918448","https://openalex.org/W4385682279","https://openalex.org/W4387839052","https://openalex.org/W4388329043","https://openalex.org/W4388886073","https://openalex.org/W4388926538","https://openalex.org/W4389713766","https://openalex.org/W4390041200","https://openalex.org/W4392678053","https://openalex.org/W4394002987","https://openalex.org/W4401042900","https://openalex.org/W4401043416","https://openalex.org/W4402670146","https://openalex.org/W4404612289","https://openalex.org/W4405181600","https://openalex.org/W4405183113","https://openalex.org/W4409150456","https://openalex.org/W4410609227","https://openalex.org/W4415250333","https://openalex.org/W4415799204","https://openalex.org/W7128378279"],"related_works":["https://openalex.org/W4412889730","https://openalex.org/W4417300637","https://openalex.org/W7202037975","https://openalex.org/W4416391310","https://openalex.org/W4417095805","https://openalex.org/W4416840250","https://openalex.org/W7165736730","https://openalex.org/W7202156404","https://openalex.org/W7130424798","https://openalex.org/W7162940118"],"abstract_inverted_index":{"Prompt":[0,18],"injection":[1,71,150],"attacks":[2],"pose":[3],"a":[4,101,108,142],"critical":[5],"threat":[6],"to":[7,36,90],"large":[8],"language":[9],"models":[10,81],"(LLMs),":[11],"enabling":[12,73],"goal":[13],"hijacking":[14],"and":[15,144,154],"data":[16],"leakage.":[17],"guard":[19,57,104],"models,":[20],"though":[21],"effective":[22],"in":[23,69],"defense,":[24],"suffer":[25,82],"from":[26,83],"over-defense":[27,53,84],"--":[28],"falsely":[29],"flagging":[30],"benign":[31,62],"inputs":[32],"as":[33],"malicious":[34],"due":[35],"trigger":[37,66,123],"word":[38],"bias.":[39],"To":[40,95],"address":[41],"this":[42],"issue,":[43],"we":[44,98],"introduce":[45],"NotInject,":[46,133],"an":[47],"evaluation":[48],"dataset":[49],"that":[50,79,106],"systematically":[51],"measures":[52],"across":[54],"various":[55],"prompt":[56,70,103,149],"models.":[58],"NotInject":[59],"contains":[60],"339":[61],"samples":[63],"enriched":[64],"with":[65,86],"words":[67],"common":[68],"attacks,":[72],"fine-grained":[74],"evaluation.":[75],"Our":[76],"results":[77],"show":[78],"state-of-the-art":[80,127],"issues,":[85],"accuracy":[87],"dropping":[88],"close":[89],"random":[91],"guessing":[92],"levels":[93],"(60%).":[94],"mitigate":[96],"this,":[97],"propose":[99],"InjecGuard,":[100],"novel":[102],"model":[105,138],"incorporates":[107],"new":[109],"training":[110],"strategy,":[111],"Mitigating":[112],"Over-defense":[113],"for":[114,147],"Free":[115],"(MOF),":[116],"which":[117],"significantly":[118],"reduces":[119],"the":[120,135],"bias":[121],"on":[122,129],"words.":[124],"InjecGuard":[125],"demonstrates":[126],"performance":[128],"diverse":[130],"benchmarks":[131],"including":[132],"surpassing":[134],"existing":[136],"best":[137],"by":[139],"30.8%,":[140],"offering":[141],"robust":[143],"open-source":[145],"solution":[146],"detecting":[148],"attacks.":[151],"The":[152],"code":[153],"datasets":[155],"are":[156],"released":[157],"at":[158],"https://github.com/leolee99/InjecGuard.":[159]},"counts_by_year":[{"year":2026,"cited_by_count":40},{"year":2025,"cited_by_count":15},{"year":2024,"cited_by_count":1}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2024-11-14T00:00:00"}