{"id":"https://openalex.org/W7162795967","doi":"https://doi.org/10.48550/arxiv.2605.28830","title":"Benchmarking Open-Source Safety Guard Models: A Comprehensive Evaluation","display_name":"Benchmarking Open-Source Safety Guard Models: A Comprehensive Evaluation","publication_year":2026,"publication_date":"2026-04-10","ids":{"openalex":"https://openalex.org/W7162795967","doi":"https://doi.org/10.48550/arxiv.2605.28830"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.28830","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28830","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,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.28830","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120289877","display_name":"Reetu Raj Harsh","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Harsh, Reetu Raj","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5040045971","display_name":"Bhaskarjit Sarmah","orcid":"https://orcid.org/0009-0004-3076-9539"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sarmah, Bhaskarjit","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137352386","display_name":"Stefano Pasquali","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Pasquali, Stefano","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/T12262","display_name":"Hate Speech and Cyberbullying Detection","score":0.33219999074935913,"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/T12262","display_name":"Hate Speech and Cyberbullying Detection","score":0.33219999074935913,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.28049999475479126,"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":0.055399999022483826,"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/guard","display_name":"Guard (computer science)","score":0.6376000046730042},{"id":"https://openalex.org/keywords/benchmarking","display_name":"Benchmarking","score":0.6086999773979187},{"id":"https://openalex.org/keywords/benchmark","display_name":"Benchmark (surveying)","score":0.42010000348091125},{"id":"https://openalex.org/keywords/metric","display_name":"Metric (unit)","score":0.4163999855518341},{"id":"https://openalex.org/keywords/safeguard","display_name":"Safeguard","score":0.3824999928474426},{"id":"https://openalex.org/keywords/moderation","display_name":"Moderation","score":0.3359000086784363},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.3246000111103058}],"concepts":[{"id":"https://openalex.org/C141141315","wikidata":"https://www.wikidata.org/wiki/Q2379942","display_name":"Guard (computer science)","level":2,"score":0.6376000046730042},{"id":"https://openalex.org/C86251818","wikidata":"https://www.wikidata.org/wiki/Q816754","display_name":"Benchmarking","level":2,"score":0.6086999773979187},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5992000102996826},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.44620001316070557},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.42260000109672546},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.42010000348091125},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.4163999855518341},{"id":"https://openalex.org/C2780771206","wikidata":"https://www.wikidata.org/wiki/Q3271761","display_name":"Safeguard","level":2,"score":0.3824999928474426},{"id":"https://openalex.org/C93225998","wikidata":"https://www.wikidata.org/wiki/Q1941972","display_name":"Moderation","level":2,"score":0.3359000086784363},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.3246000111103058},{"id":"https://openalex.org/C111219384","wikidata":"https://www.wikidata.org/wiki/Q6954384","display_name":"NIST","level":2,"score":0.31869998574256897},{"id":"https://openalex.org/C45804977","wikidata":"https://www.wikidata.org/wiki/Q7239673","display_name":"Predictive modelling","level":2,"score":0.30799999833106995},{"id":"https://openalex.org/C2992838255","wikidata":"https://www.wikidata.org/wiki/Q856887","display_name":"Security guard","level":2,"score":0.30219998955726624},{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.29019999504089355},{"id":"https://openalex.org/C12174686","wikidata":"https://www.wikidata.org/wiki/Q1058438","display_name":"Risk assessment","level":2,"score":0.28189998865127563},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.2711000144481659},{"id":"https://openalex.org/C32896092","wikidata":"https://www.wikidata.org/wiki/Q189447","display_name":"Risk management","level":2,"score":0.2676999866962433}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.28830","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28830","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,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.28830","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.28830","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,"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"},"sustainable_development_goals":[{"display_name":"Gender equality","id":"https://metadata.un.org/sdg/5","score":0.7494385242462158}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"As":[0],"Large":[1],"Language":[2],"Models":[3],"(LLMs)":[4],"are":[5],"increasingly":[6],"deployed":[7],"in":[8,158],"safety-critical":[9],"applications,":[10,82],"robust":[11],"content":[12,59,86],"moderation":[13],"becomes":[14],"essential.":[15],"We":[16,72,128],"present":[17],"a":[18,28],"comprehensive":[19],"evaluation":[20,94],"of":[21,31,125],"14":[22],"open-source":[23],"safety":[24,40,81,137,155],"guard":[25,143,156],"models":[26,109,144,157],"on":[27,57],"curated":[29],"benchmark":[30,43],"79,331":[32],"samples":[33],"spanning":[34],"8":[35],"NIST":[36],"AI":[37],"Risk":[38],"Framework":[39],"categories.":[41],"Our":[42,93],"aggregates":[44],"four":[45],"diverse":[46],"datasets":[47],"(HarmBench,":[48],"StrongREJECT,":[49],"RealToxicityPrompts,":[50],"and":[51,69,114,140],"BeaverTails),":[52],"filtered":[53],"to":[54,123],"focus":[55],"exclusively":[56],"safety-relevant":[58],"(violence,":[60],"hate":[61],"speech,":[62],"harassment,":[63],"sexual":[64],"content,":[65],"suicide/self-harm,":[66],"profanity,":[67],"threats,":[68],"health":[70],"misinformation).":[71],"find":[73],"that":[74,130,141],"recall":[75,105],"is":[76],"the":[77,103],"critical":[78],"metric":[79],"for":[80,153],"as":[83],"missing":[84,121],"harmful":[85],"poses":[87],"greater":[88],"risk":[89],"than":[90],"false":[91],"positives.":[92],"reveals":[95],"surprising":[96],"results:":[97],"Qwen":[98],"Guard":[99,112],"(4B":[100],"parameters)":[101],"achieves":[102],"highest":[104],"(83.97%)":[106],"while":[107],"larger":[108],"like":[110],"Llama":[111],"(12B)":[113],"GPT-OSS":[115],"Safeguard":[116],"(20B)":[117],"exhibit":[118],"conservative":[119],"behavior,":[120],"up":[122],"75%":[124],"unsafe":[126],"content.":[127],"demonstrate":[129],"model":[131],"size":[132],"does":[133],"not":[134],"correlate":[135],"with":[136],"detection":[138],"performance":[139],"general-purpose":[142],"outperform":[145],"specialized":[146],"ones.":[147],"These":[148],"findings":[149],"provide":[150],"practical":[151],"guidance":[152],"selecting":[154],"production":[159],"deployments.":[160]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-30T00:00:00"}