{"id":"https://openalex.org/W7165752568","doi":"https://doi.org/10.48550/arxiv.2606.24245","title":"AutoSpec: Safety Rule Evolution for LLM Agents via Inductive Logic Programming","display_name":"AutoSpec: Safety Rule Evolution for LLM Agents via Inductive Logic Programming","publication_year":2026,"publication_date":"2026-06-23","ids":{"openalex":"https://openalex.org/W7165752568","doi":"https://doi.org/10.48550/arxiv.2606.24245"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.24245","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.24245","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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.2606.24245","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139231681","display_name":"Pingchuan Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Pingchuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139291031","display_name":"Zhaoyu Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zhaoyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124125427","display_name":"Zimo Ji","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ji, Zimo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139248552","display_name":"Yuguang Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Yuguang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5120641627","display_name":"Zhantong Xue","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xue, Zhantong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139235794","display_name":"Zongjie Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Zongjie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139253527","display_name":"Shuai Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Shuai","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139240089","display_name":"Xiaoqin Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Xiaoqin","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/T10028","display_name":"Topic Modeling","score":0.21719999611377716,"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/T10028","display_name":"Topic Modeling","score":0.21719999611377716,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.15719999372959137,"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":0.14980000257492065,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"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/interpretability","display_name":"Interpretability","score":0.7924000024795532},{"id":"https://openalex.org/keywords/inductive-logic-programming","display_name":"Inductive logic programming","score":0.7113999724388123},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.6577000021934509},{"id":"https://openalex.org/keywords/false-positive-paradox","display_name":"False positive paradox","score":0.572700023651123},{"id":"https://openalex.org/keywords/heuristic","display_name":"Heuristic","score":0.5361999869346619},{"id":"https://openalex.org/keywords/false-positives-and-false-negatives","display_name":"False positives and false negatives","score":0.43140000104904175},{"id":"https://openalex.org/keywords/domain","display_name":"Domain (mathematical analysis)","score":0.4065000116825104},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.3926999866962433},{"id":"https://openalex.org/keywords/heuristics","display_name":"Heuristics","score":0.37279999256134033}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.7924000024795532},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7570000290870667},{"id":"https://openalex.org/C2779382394","wikidata":"https://www.wikidata.org/wiki/Q1464197","display_name":"Inductive logic programming","level":2,"score":0.7113999724388123},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6577000021934509},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5813999772071838},{"id":"https://openalex.org/C64869954","wikidata":"https://www.wikidata.org/wiki/Q1859747","display_name":"False positive paradox","level":2,"score":0.572700023651123},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.5361999869346619},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4993000030517578},{"id":"https://openalex.org/C112789634","wikidata":"https://www.wikidata.org/wiki/Q18207010","display_name":"False positives and false negatives","level":3,"score":0.43140000104904175},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.4065000116825104},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3926999866962433},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.37279999256134033},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.36890000104904175},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.36629998683929443},{"id":"https://openalex.org/C149271511","wikidata":"https://www.wikidata.org/wiki/Q1417149","display_name":"Rule-based system","level":2,"score":0.32850000262260437},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.31360000371932983},{"id":"https://openalex.org/C28180684","wikidata":"https://www.wikidata.org/wiki/Q4080983","display_name":"Memory safety","level":3,"score":0.3100999891757965},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.3068000078201294},{"id":"https://openalex.org/C207685749","wikidata":"https://www.wikidata.org/wiki/Q2088941","display_name":"Domain knowledge","level":2,"score":0.298799991607666},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.29159998893737793},{"id":"https://openalex.org/C3746660","wikidata":"https://www.wikidata.org/wiki/Q1068763","display_name":"Rule of inference","level":2,"score":0.289000004529953},{"id":"https://openalex.org/C159718280","wikidata":"https://www.wikidata.org/wiki/Q5526353","display_name":"Gas meter prover","level":3,"score":0.2768000066280365},{"id":"https://openalex.org/C144745244","wikidata":"https://www.wikidata.org/wiki/Q4927286","display_name":"Blocking (statistics)","level":2,"score":0.274399995803833},{"id":"https://openalex.org/C128838566","wikidata":"https://www.wikidata.org/wiki/Q275603","display_name":"Logic programming","level":2,"score":0.2605000138282776},{"id":"https://openalex.org/C31170391","wikidata":"https://www.wikidata.org/wiki/Q188619","display_name":"Hierarchy","level":2,"score":0.2596000134944916},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.25859999656677246}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.24245","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.24245","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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.2606.24245","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.24245","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":[{"id":"https://metadata.un.org/sdg/16","score":0.49128931760787964,"display_name":"Peace, Justice and strong institutions"},{"id":"https://metadata.un.org/sdg/10","score":0.46059346199035645,"display_name":"Reduced inequalities"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1,11],"model":[2],"(LLM)":[3],"agents":[4,25],"increasingly":[5],"automate":[6],"complex":[7],"tasks":[8],"by":[9,100],"integrating":[10],"models":[12],"with":[13,50],"external":[14],"tools":[15],"and":[16,110,122,137,187,199,209,226,249],"environments.":[17],"However,":[18],"their":[19],"autonomy":[20],"poses":[21],"significant":[22],"safety":[23,38,88],"risks:":[24],"may":[26],"execute":[27],"destructive":[28],"commands,":[29],"leak":[30],"sensitive":[31],"data,":[32],"or":[33],"violate":[34],"domain":[35],"constraints.":[36],"Existing":[37],"approaches":[39],"face":[40],"a":[41,81,111],"fundamental":[42],"tradeoff:":[43],"hand-crafted":[44],"rules":[45,53,62,89,109,183,245],"are":[46,246],"interpretable":[47,182],"but":[48,160],"brittle,":[49],"overly":[51],"conservative":[52],"blocking":[54],"safe":[55],"operations":[56],"(high":[57,66],"false":[58,67,158,163,219],"positives)":[59],"while":[60,222],"permissive":[61],"miss":[63],"unsafe":[64],"behaviors":[65],"negatives).":[68],"Neural":[69],"classifiers":[70],"lack":[71],"the":[72,107,142,170,212],"interpretability":[73],"required":[74],"for":[75],"safety-critical":[76],"deployments.":[77],"We":[78,189],"present":[79],"AutoSpec,":[80],"framework":[82],"that":[83,149,154,184],"automatically":[84],"evolves":[85],"deployed":[86],"expert-designed":[87],"from":[90,106],"user":[91],"safe/unsafe":[92],"annotations":[93],"through":[94],"counterexample-guided":[95],"inductive":[96,101],"synthesis":[97],"(CEGIS)":[98],"guided":[99],"logic":[102],"programming":[103],"(ILP).":[104],"Starting":[105],"expert":[108],"stream":[112],"of":[113,174],"annotated":[114],"traces,":[115],"AutoSpec":[116,191,203],"iteratively":[117],"evaluates":[118],"rules,":[119],"mines":[120],"false-positive":[121],"false-negative":[123],"counterexamples,":[124],"uses":[125],"ILP":[126,150],"to":[127,140,207,217,236,251],"learn":[128],"which":[129],"predicates":[130,153],"discriminate":[131],"them,":[132],"generates":[133],"candidate":[134],"rule":[135,175,205],"edits,":[136],"verifies":[138],"candidates":[139],"select":[141],"best":[143],"revision.":[144],"The":[145,231,243],"key":[146],"insight":[147],"is":[148],"efficiently":[151],"identifies":[152],"appear":[155],"frequently":[156],"in":[157,162],"negatives":[159],"rarely":[161],"positives":[164],"(or":[165],"vice":[166],"versa),":[167],"dramatically":[168],"pruning":[169],"exponential":[171],"search":[172],"space":[173],"edits.":[176],"This":[177],"continues":[178],"until":[179],"convergence,":[180],"producing":[181],"balance":[185],"precision":[186],"recall.":[188],"evaluate":[190],"on":[192],"291":[193],"execution":[194,198],"traces":[195],"spanning":[196],"code":[197],"embodied":[200],"agent":[201],"domains.":[202],"raises":[204],"F1":[206,239],"0.98":[208],"0.93":[210],"across":[211],"two":[213],"domains,":[214],"achieving":[215],"up":[216,235],"94%":[218],"positive":[220],"reduction":[221],"maintaining":[223],"high":[224],"recall,":[225],"converges":[227],"within":[228],"4-5":[229],"iterations.":[230],"ILP-guided":[232],"approach":[233],"achieves":[234],"4.8x":[237],"higher":[238],"than":[240],"heuristic":[241],"CEGIS.":[242],"learned":[244],"human-readable,":[247],"auditable,":[248],"generalize":[250],"unseen":[252],"scenarios.":[253]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-25T00:00:00"}