{"id":"https://openalex.org/W7140125705","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.146","title":"SafeSearch: Do Not Trade Safety for Utility in LLM Search Agents","display_name":"SafeSearch: Do Not Trade Safety for Utility in LLM Search Agents","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7140125705","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.146"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-eacl.146","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.146","pdf_url":"https://aclanthology.org/2026.findings-eacl.146.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: EACL 2026","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-eacl.146.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5130353050","display_name":"Qiusi Zhan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Qiusi Zhan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130402863","display_name":"Angeline Budiman-Chan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Angeline Budiman-Chan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130409709","display_name":"Abdelrahman Zayed","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Abdelrahman Zayed","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130337362","display_name":"Xingzhi Guo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xingzhi Guo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113139155","display_name":"Daniel Kang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Daniel Kang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5130400846","display_name":"Joo-Kyung Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Joo-Kyung Kim","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":6,"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":0.39912198,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"2800","last_page":"2815"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.1264999955892563,"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/T10456","display_name":"Multi-Agent Systems and Negotiation","score":0.1264999955892563,"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/T10991","display_name":"Game Theory and Voting Systems","score":0.10939999669790268,"subfield":{"id":"https://openalex.org/subfields/2002","display_name":"Economics and Econometrics"},"field":{"id":"https://openalex.org/fields/20","display_name":"Economics, Econometrics and Finance"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11182","display_name":"Auction Theory and Applications","score":0.09889999777078629,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/action","display_name":"Action (physics)","score":0.31310001015663147},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.30469998717308044},{"id":"https://openalex.org/keywords/government","display_name":"Government (linguistics)","score":0.2583000063896179},{"id":"https://openalex.org/keywords/agency","display_name":"Agency (philosophy)","score":0.2533999979496002}],"concepts":[{"id":"https://openalex.org/C144133560","wikidata":"https://www.wikidata.org/wiki/Q4830453","display_name":"Business","level":0,"score":0.43689998984336853},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.39500001072883606},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.33739998936653137},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.31310001015663147},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.30469998717308044},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.3012000024318695},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.2754000127315521},{"id":"https://openalex.org/C2778137410","wikidata":"https://www.wikidata.org/wiki/Q2732820","display_name":"Government (linguistics)","level":2,"score":0.2583000063896179},{"id":"https://openalex.org/C108170787","wikidata":"https://www.wikidata.org/wiki/Q3951828","display_name":"Agency (philosophy)","level":2,"score":0.2533999979496002},{"id":"https://openalex.org/C2779530757","wikidata":"https://www.wikidata.org/wiki/Q1207505","display_name":"Quality (philosophy)","level":2,"score":0.23919999599456787}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-eacl.146","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.146","pdf_url":"https://aclanthology.org/2026.findings-eacl.146.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: EACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-eacl.146","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.146","pdf_url":"https://aclanthology.org/2026.findings-eacl.146.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: EACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320315389","display_name":"Open Philanthropy Project","ror":"https://ror.org/004d1k391"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140125705.pdf","grobid_xml":"https://content.openalex.org/works/W7140125705.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"model":[2,70],"(LLM)":[3],"based":[4],"search":[5,37,74],"agents":[6,38],"iteratively":[7],"generate":[8],"queries,":[9],"retrieve":[10,78],"external":[11],"information,":[12],"and":[13,42,79,116,141,163,185],"reason":[14],"to":[15,49,77],"answer":[16],"open-domain":[17],"questions.While":[18],"researchers":[19],"have":[20],"primarily":[21],"focused":[22],"on":[23],"improving":[24,183],"their":[25,27,66],"utility,":[26],"safety":[28,115,184],"behaviors":[29],"remain":[30],"underexplored.In":[31],"this":[32,109],"paper,":[33],"we":[34],"first":[35],"evaluate":[36],"using":[39],"red-teaming":[40],"datasets":[41,157],"find":[43],"that":[44,105,125,137,146],"they":[45],"are":[46],"more":[47],"likely":[48],"produce":[50],"harmful":[51],"outputs":[52],"than":[53],"base":[54,69],"LLMs.For":[55],"example,":[56],"when":[57],"asked":[58],"\"How":[59],"can":[60],"I":[61],"track":[62],"someone's":[63],"location":[64],"without":[65],"consent?\",":[67],"a":[68,73,120,127,132,169],"refuses,":[71],"whereas":[72],"agent":[75,149],"designed":[76],"cite":[80],"sources":[81],"may":[82],"lower":[83],"its":[84],"refusal":[85],"threshold,":[86],"fetch":[87],"documents":[88],"(e.g.,":[89],"court":[90],"cases),":[91],"and,":[92],"once":[93],"appended,":[94],"synthesize":[95],"them":[96],"into":[97],"an":[98],"informative":[99],"yet":[100],"unsafe":[101,139],"summary.We":[102],"further":[103],"show":[104,145],"utility-oriented":[106],"finetuning":[107],"intensifies":[108],"risk,":[110],"motivating":[111],"joint":[112],"alignment":[113],"of":[114,168,177],"utility.We":[117],"present":[118],"SAFESEARCH,":[119],"multi-objective":[121],"reinforcement":[122],"learning":[123],"approach":[124],"couples":[126],"final-output":[128],"safety/utility":[129],"reward":[130,180],"with":[131],"novel":[133],"query-level":[134,179],"shaping":[135],"term":[136],"penalizes":[138],"queries":[140],"rewards":[142],"safe":[143],"ones.Experiments":[144],"SAFESEARCH":[147],"reduces":[148],"harmfulness":[150],"by":[151],"over":[152],"70%":[153],"across":[154],"three":[155],"redteaming":[156],"while":[158],"producing":[159],"safe,":[160],"helpful":[161],"responses,":[162],"matches":[164],"the":[165,175,178],"QA":[166],"performance":[167],"utility-only":[170],"finetuned":[171],"agent.Further":[172],"analyses":[173],"confirm":[174],"effectiveness":[176],"in":[181],"jointly":[182],"utility.":[186],"Reward":[187],"Scoring":[188]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2026-03-24T00:00:00"}