{"id":"https://openalex.org/W7164025639","doi":"https://doi.org/10.48550/arxiv.2606.09165","title":"Reliable to Expressive: A Curriculum for Rubric-Following Safety Judges","display_name":"Reliable to Expressive: A Curriculum for Rubric-Following Safety Judges","publication_year":2026,"publication_date":"2026-06-08","ids":{"openalex":"https://openalex.org/W7164025639","doi":"https://doi.org/10.48550/arxiv.2606.09165"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.09165","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.09165","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.2606.09165","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5102079663","display_name":"Yongtaek Lim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lim, Yongtaek","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138277827","display_name":"Hyeji Choi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Choi, Hyeji","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138254583","display_name":"Minwoo Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kim, Minwoo","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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.3797999918460846,"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":0.3797999918460846,"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":0.13420000672340393,"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":0.06589999794960022,"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/rubric","display_name":"Rubric","score":0.9667999744415283},{"id":"https://openalex.org/keywords/curriculum","display_name":"Curriculum","score":0.6675999760627747},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.5317999720573425},{"id":"https://openalex.org/keywords/schedule","display_name":"Schedule","score":0.38989999890327454},{"id":"https://openalex.org/keywords/variance","display_name":"Variance (accounting)","score":0.35920000076293945},{"id":"https://openalex.org/keywords/baseline","display_name":"Baseline (sea)","score":0.29179999232292175}],"concepts":[{"id":"https://openalex.org/C111640148","wikidata":"https://www.wikidata.org/wiki/Q847349","display_name":"Rubric","level":2,"score":0.9667999744415283},{"id":"https://openalex.org/C47177190","wikidata":"https://www.wikidata.org/wiki/Q207137","display_name":"Curriculum","level":2,"score":0.6675999760627747},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5317999720573425},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5266000032424927},{"id":"https://openalex.org/C68387754","wikidata":"https://www.wikidata.org/wiki/Q7271585","display_name":"Schedule","level":2,"score":0.38989999890327454},{"id":"https://openalex.org/C145420912","wikidata":"https://www.wikidata.org/wiki/Q853077","display_name":"Mathematics education","level":1,"score":0.3693999946117401},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.35920000076293945},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.3555999994277954},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.29179999232292175},{"id":"https://openalex.org/C2779137570","wikidata":"https://www.wikidata.org/wiki/Q16243196","display_name":"EXPOSE","level":2,"score":0.29179999232292175},{"id":"https://openalex.org/C2776372474","wikidata":"https://www.wikidata.org/wiki/Q508291","display_name":"Simplicity","level":2,"score":0.2883000075817108},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.2849999964237213},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.28220000863075256},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.2734000086784363},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.2709999978542328}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.09165","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.09165","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.2606.09165","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.09165","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":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Safety":[0],"judges":[1,153],"are":[2],"increasingly":[3],"deployed":[4],"to":[5,31,82,86,155],"evaluate":[6,110],"model":[7],"outputs":[8],"against":[9],"evolving":[10],"criteria,":[11,91],"yet":[12],"recent":[13],"meta-evaluation":[14],"work":[15],"shows":[16,165],"they":[17],"remain":[18],"brittle":[19],"under":[20,116],"prompt":[21],"and":[22,92,103,123,151,161,185],"rubric":[23,59,119,175,190],"variation,":[24],"with":[25,99,138],"false":[26],"negative-rate":[27],"swings":[28],"of":[29,89,142],"up":[30,154],"0.24":[32],"reported":[33],"for":[34],"stylistic":[35],"perturbations":[36],"alone.":[37],"We":[38,67,109],"argue":[39],"that":[40,72,97,166],"safety":[41,149],"judgment":[42],"is":[43],"fundamentally":[44],"a":[45,48,69,94,112,124,139],"rubric-following":[46],"problem:":[47],"robust":[49],"judge":[50,85,130],"must":[51],"apply":[52],"the":[53,84,87,135,181,188],"given":[54],"evaluation":[55,90],"criteria":[56],"consistently":[57],"across":[58,134],"formulations":[60],"rather":[61],"than":[62],"memorize":[63],"one":[64],"specific":[65],"template.":[66],"propose":[68],"training":[70],"strategy":[71],"combines":[73],"(i)":[74],"instance-conditioned":[75],"dynamic":[76,169],"rubrics":[77,137,170],"generated":[78],"from":[79],"prompt-response-label":[80],"triples":[81],"expose":[83],"variability":[88],"(ii)":[93],"reliable-to-expressive":[95],"curriculum":[96,129,182],"begins":[98],"clean":[100],"fixed-rubric":[101],"supervision":[102],"progressively":[104],"introduces":[105],"noisier":[106],"dynamic-rubric":[107],"data.":[108],"on":[111,187],"single":[113],"human-labeled":[114],"set":[115],"three":[117,136],"contrasting":[118],"prompts":[120],"(HarmBench-style,":[121],"ShieldGemma-style,":[122],"domain-specific":[125],"rubric).":[126],"Our":[127],"12B":[128],"achieves":[131],"94.12-94.88%":[132],"accuracy":[133,160],"cross-rubric":[140],"range":[141],"only":[143,180],"0.76,":[144],"outperforming":[145],"general-purpose":[146],"LLMs,":[147],"dedicated":[148],"classifiers,":[150],"reasoning-oriented":[152],"30B":[156],"in":[157],"both":[158],"peak":[159],"stability.":[162],"An":[163],"ablation":[164],"naively":[167],"mixing":[168],"into":[171],"SFT":[172],"increases":[173],"cross":[174],"variance":[176],"(1.44":[177],"->":[178],"3.60);":[179],"schedule":[183],"recovers":[184],"improves":[186],"fixed":[189],"baseline":[191],"(variance":[192],"0.76).":[193]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-10T00:00:00"}