{"id":"https://openalex.org/W4400480697","doi":"https://doi.org/10.48550/arxiv.2407.05557","title":"$R^2$-Guard: Robust Reasoning Enabled LLM Guardrail via Knowledge-Enhanced Logical Reasoning","display_name":"$R^2$-Guard: Robust Reasoning Enabled LLM Guardrail via Knowledge-Enhanced Logical Reasoning","publication_year":2024,"publication_date":"2024-07-08","ids":{"openalex":"https://openalex.org/W4400480697","doi":"https://doi.org/10.48550/arxiv.2407.05557"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2407.05557","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.05557","pdf_url":"https://arxiv.org/pdf/2407.05557","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","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/2407.05557","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5029251926","display_name":"Mintong Kang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kang, Mintong","raw_affiliation_strings":["UIUC"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UIUC","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"last","author":{"id":null,"display_name":"Li, Bo","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40347166","display_name":"University of Chicago","ror":"https://ror.org/024mw5h28","country_code":"US","type":"education","lineage":["https://openalex.org/I40347166"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Li, Bo","raw_affiliation_strings":["UIUC & Uchicago"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UIUC & Uchicago","institution_ids":["https://openalex.org/I40347166"]}]}],"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":35,"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":0.9223999977111816,"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":0.9223999977111816,"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":0.04740000143647194,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.010099999606609344,"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/llm-guardrails","display_name":"LLM guardrails","score":0.8870000243186951},{"id":"https://openalex.org/keywords/large-language-model-safety","display_name":"large language model safety","score":0.7900000214576721},{"id":"https://openalex.org/keywords/jailbreak-attacks","display_name":"jailbreak attacks","score":0.675000011920929},{"id":"https://openalex.org/keywords/probabilistic-graphical-models","display_name":"probabilistic graphical models","score":0.7440000176429749},{"id":"https://openalex.org/keywords/markov-logic-networks","display_name":"Markov logic networks","score":0.9750000238418579},{"id":"https://openalex.org/keywords/probabilistic-circuits","display_name":"probabilistic circuits","score":1.0}],"concepts":[{"id":"https://openalex.org/C141141315","wikidata":"https://www.wikidata.org/wiki/Q2379942","display_name":"Guard (computer science)","level":2,"score":0.7100827693939209},{"id":"https://openalex.org/C97364631","wikidata":"https://www.wikidata.org/wiki/Q484284","display_name":"Deductive reasoning","level":2,"score":0.5813153982162476},{"id":"https://openalex.org/C37335422","wikidata":"https://www.wikidata.org/wiki/Q6888134","display_name":"Model-based reasoning","level":3,"score":0.552593469619751},{"id":"https://openalex.org/C86827895","wikidata":"https://www.wikidata.org/wiki/Q7098582","display_name":"Opportunistic reasoning","level":4,"score":0.5168745517730713},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.49208563566207886},{"id":"https://openalex.org/C43971567","wikidata":"https://www.wikidata.org/wiki/Q3142865","display_name":"Logical reasoning","level":2,"score":0.4790145754814148},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.45300307869911194},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3771568536758423},{"id":"https://openalex.org/C161301231","wikidata":"https://www.wikidata.org/wiki/Q3478658","display_name":"Knowledge representation and reasoning","level":2,"score":0.24795866012573242},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.10539031028747559}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2407.05557","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.05557","pdf_url":"https://arxiv.org/pdf/2407.05557","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2407.05557","doi":"https://doi.org/10.48550/arxiv.2407.05557","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2407.05557","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":"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":"pmh:oai:arXiv.org:2407.05557","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.05557","pdf_url":"https://arxiv.org/pdf/2407.05557","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G2882411384","display_name":"AF: Small: Collaborative Research: Rigorous Approaches for Scalable Privacy-preserving Deep Learning","funder_award_id":"1910100","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G6197153243","display_name":null,"funder_award_id":"2229876","funder_id":"https://openalex.org/F4320332180","funder_display_name":"Defense Advanced Research Projects Agency"},{"id":"https://openalex.org/G6360409883","display_name":"AI Institute for Agent-based Cyber Threat Intelligence and Operation","funder_award_id":"2229876","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G6860822721","display_name":null,"funder_award_id":"80NSSC20M0229","funder_id":"https://openalex.org/F4320306101","funder_display_name":"National Aeronautics and Space Administration"},{"id":"https://openalex.org/G7894451278","display_name":"CAREER: DeepTrust: Enabling Robust Machine Learning with Exogenous Information","funder_award_id":"2046726","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320306101","display_name":"National Aeronautics and Space Administration","ror":"https://ror.org/027ka1x80"},{"id":"https://openalex.org/F4320322037","display_name":"Nuclear Safety and Security Commission","ror":"https://ror.org/05qk3ge34"},{"id":"https://openalex.org/F4320332180","display_name":"Defense Advanced Research Projects Agency","ror":"https://ror.org/02caytj08"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4400480697.pdf","grobid_xml":"https://content.openalex.org/works/W4400480697.grobid-xml"},"referenced_works_count":32,"referenced_works":["https://openalex.org/W2427182019","https://openalex.org/W2908242605","https://openalex.org/W2963533931","https://openalex.org/W3201425170","https://openalex.org/W4366850566","https://openalex.org/W4385374425","https://openalex.org/W4386185625","https://openalex.org/W4386228564","https://openalex.org/W4386436506","https://openalex.org/W4386907292","https://openalex.org/W4387430709","https://openalex.org/W4387559087","https://openalex.org/W4387725416","https://openalex.org/W4389518965","https://openalex.org/W4389713766","https://openalex.org/W4390601403","https://openalex.org/W4391463145","https://openalex.org/W4391632220","https://openalex.org/W4392616695","https://openalex.org/W4392972201","https://openalex.org/W4393064055","https://openalex.org/W4394709004","https://openalex.org/W4398229667","https://openalex.org/W4399205773","https://openalex.org/W4401042900","https://openalex.org/W4402670748","https://openalex.org/W4402671779","https://openalex.org/W4403511268","https://openalex.org/W4405181600","https://openalex.org/W4410609100","https://openalex.org/W4415795355","https://openalex.org/W4415797457"],"related_works":["https://openalex.org/W4407012568","https://openalex.org/W7166821409","https://openalex.org/W7169856787","https://openalex.org/W6892027301","https://openalex.org/W7162814911","https://openalex.org/W4412888090","https://openalex.org/W7202293563","https://openalex.org/W7125484780","https://openalex.org/W7117264240","https://openalex.org/W7090908468"],"abstract_inverted_index":{"As":[0],"LLMs":[1],"become":[2],"increasingly":[3],"prevalent":[4],"across":[5],"various":[6,25],"applications,":[7],"it":[8],"is":[9],"critical":[10],"to":[11,15,31,41,47,58,145,169,191],"establish":[12],"safety":[13,26,55,65,107,112,195,217],"guardrails":[14],"moderate":[16],"input/output":[17],"content":[18,104],"of":[19,102,136,156,206,223],"LLMs.":[20],"Existing":[21],"guardrail":[22,79,97,141,183,213],"models":[23,98,142,214],"treat":[24],"categories":[27,116,138],"independently":[28],"and":[29,61,92,121,162,166,219,240],"fail":[30],"explicitly":[32],"capture":[33],"the":[34,146,204,221],"intercorrelations":[35],"among":[36,114],"them.":[37],"This":[38],"has":[39],"led":[40],"limitations":[42],"such":[43],"as":[44,117],"ineffectiveness":[45],"due":[46],"inadequate":[48],"training":[49],"on":[50,105,215,238],"long-tail":[51],"data":[52],"from":[53,139],"correlated":[54],"categories,":[56],"susceptibility":[57],"jailbreaking":[59,228,244],"attacks,":[60],"inflexibility":[62],"regarding":[63],"new":[64,194],"categories.":[66,108,201],"To":[67,177],"address":[68],"these":[69],"limitations,":[70],"we":[71,185],"propose":[72],"$R^2$-Guard,":[73],"a":[74,125,187,193],"robust":[75],"reasoning":[76,93,131,147],"enabled":[77],"LLM":[78],"via":[80,173],"knowledge-enhanced":[81],"logical":[82,119],"reasoning.":[83],"Specifically,":[84],"$R^2$-Guard":[85,207,224,230],"comprises":[86],"two":[87,154],"parts:":[88],"data-driven":[89,96,140],"category-specific":[90],"learning":[91],"components.":[94],"The":[95,133],"provide":[99],"unsafety":[100,134],"probabilities":[101,135],"moderated":[103],"different":[106,115,137],"We":[109,152,202],"then":[110],"encode":[111],"knowledge":[113],"first-order":[118],"rules":[120],"embed":[122],"them":[123],"into":[124],"probabilistic":[126,163],"graphic":[127],"model":[128],"(PGM)":[129],"based":[130],"component.":[132],"are":[143],"sent":[144],"component":[148],"for":[149,182],"final":[150],"inference.":[151],"employ":[153,186],"types":[155],"PGMs:":[157],"Markov":[158],"logic":[159],"networks":[160],"(MLNs)":[161],"circuits":[164],"(PCs),":[165],"optimize":[167],"PCs":[168],"achieve":[170],"precision-efficiency":[171],"balance":[172],"improved":[174],"graph":[175],"structure.":[176],"further":[178],"perform":[179],"stress":[180],"tests":[181],"models,":[184],"pairwise":[188],"construction":[189],"method":[190,234],"construct":[192],"benchmark":[196],"TwinSafety,":[197],"which":[198],"features":[199],"principled":[200],"demonstrate":[203,220],"effectiveness":[205],"by":[208,236,241],"comparisons":[209],"with":[210],"eight":[211],"strong":[212],"six":[216],"benchmarks,":[218],"robustness":[222],"against":[225,243],"four":[226],"SOTA":[227,233],"attacks.":[229,245],"significantly":[231],"surpasses":[232],"LlamaGuard":[235],"30.2%":[237],"ToxicChat":[239],"59.5%":[242]},"counts_by_year":[{"year":2026,"cited_by_count":16},{"year":2025,"cited_by_count":18},{"year":2024,"cited_by_count":1}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2024-07-10T00:00:00"}