{"id":"https://openalex.org/W7166831561","doi":"https://doi.org/10.18653/v1/2026.findings-acl.513","title":"Detecting Proxy Gaming in RL and LLM Alignment via Evaluator Stress Tests","display_name":"Detecting Proxy Gaming in RL and LLM Alignment via Evaluator Stress Tests","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166831561","doi":"https://doi.org/10.18653/v1/2026.findings-acl.513"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.513","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.513","pdf_url":"https://aclanthology.org/2026.findings-acl.513.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.513.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5119832027","display_name":"Sanjeda Akter","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sanjeda Akter","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139844005","display_name":"Ibne Farabi Shihab","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ibne Farabi Shihab","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139733942","display_name":"Anuj Sharma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Anuj Sharma","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":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.84823395,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"10554","last_page":"10583"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11644","display_name":"Spam and Phishing Detection","score":0.0778999999165535,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T11644","display_name":"Spam and Phishing Detection","score":0.0778999999165535,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/T11652","display_name":"Imbalanced Data Classification Techniques","score":0.07580000162124634,"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/T11598","display_name":"Internet Traffic Analysis and Secure E-voting","score":0.06629999727010727,"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/proxy","display_name":"Proxy (statistics)","score":0.428600013256073},{"id":"https://openalex.org/keywords/stress","display_name":"Stress (linguistics)","score":0.2824000120162964},{"id":"https://openalex.org/keywords/stress-testing","display_name":"Stress testing (software)","score":0.2624000012874603},{"id":"https://openalex.org/keywords/statistical-hypothesis-testing","display_name":"Statistical hypothesis testing","score":0.2379000037908554}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5428000092506409},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.428600013256073},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.39430001378059387},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3490999937057495},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2906999886035919},{"id":"https://openalex.org/C21036866","wikidata":"https://www.wikidata.org/wiki/Q181767","display_name":"Stress (linguistics)","level":2,"score":0.2824000120162964},{"id":"https://openalex.org/C7515471","wikidata":"https://www.wikidata.org/wiki/Q1936882","display_name":"Stress testing (software)","level":2,"score":0.2624000012874603},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.2572999894618988},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.2554999887943268},{"id":"https://openalex.org/C87007009","wikidata":"https://www.wikidata.org/wiki/Q210832","display_name":"Statistical hypothesis testing","level":2,"score":0.2379000037908554}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.513","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.513","pdf_url":"https://aclanthology.org/2026.findings-acl.513.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.513","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.513","pdf_url":"https://aclanthology.org/2026.findings-acl.513.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166831561.pdf","grobid_xml":"https://content.openalex.org/works/W7166831561.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Proxy":[0],"optimization,":[1],"where":[2],"AI":[3],"systems":[4],"exploit":[5],"evaluator":[6],"weaknesses":[7],"rather":[8],"than":[9],"improve":[10],"intended":[11],"objectives,":[12],"threatens":[13],"both":[14,58,141],"reinforcement":[15],"learning":[16],"(reward":[17],"hacking)":[18],"and":[19,64,74,88,98,132,146],"LLM":[20,77,148],"alignment":[21],"(evaluator":[22],"gaming).We":[23],"introduce":[24],"the":[25],"Evaluator":[26],"Stress":[27],"Test":[28],"(EST),":[29],"an":[30],"invariancebased":[31],"framework":[32],"that":[33,110],"detects":[34],"proxy":[35],"gaming":[36],"by":[37,128,135],"separating":[38],"exploitable":[39],"sensitivity":[40],"(e.g.,":[41],"format,":[42],"physics":[43],"bugs)":[44],"from":[45],"content-driven":[46],"improvements":[47],"using":[48],"controlled":[49],"perturbations":[50],"with":[51,101],"semantic":[52],"validity":[53],"audits.We":[54],"validate":[55],"EST":[56,70,94],"across":[57,61,79],"domains.In":[59],"RL,":[60],"15":[62],"environments":[63],"5":[65],"algorithms":[66],"(2,156":[67],"expertannotated":[68],"episodes),":[69],"achieves":[71,95],"78.4%":[72],"precision":[73,97],"81.7%":[75],"recall.In":[76],"alignment,":[78],"4":[80],"tasks,":[81],"2":[82,85,89],"model":[83],"scales,":[84],"training":[86],"methods,":[87],"judges":[90],"(1,200":[91],"human-annotated":[92],"instances),":[93],"74.2%":[96],"78.6%":[99],"recall":[100],"early":[102],"warning":[103],"signals":[104],"preceding":[105],"quality":[106],"decline.Cross-domain":[107],"analysis":[108],"reveals":[109],"proxy-true":[111],"correlation":[112],"tracking":[113],"transfers":[114],"directly":[115],"between":[116],"domains,":[117],"while":[118],"perturbation":[119],"design":[120],"requires":[121],"domain":[122],"adaptation.Closed-loop":[123],"mitigation":[124],"improves":[125],"human":[126],"winrate":[127],"8.3":[129],"points":[130],"(LLM)":[131],"reduces":[133],"hacking":[134],"54.6%":[136],"(RL).We":[137],"release":[138],"benchmarks":[139],"for":[140],"domains:":[142],"2,156":[143],"RL":[144],"episodes":[145],"1,200":[147],"instances.":[149]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}