{"id":"https://openalex.org/W7220698552","doi":"https://doi.org/10.48550/arxiv.2610.04409","title":"Understanding and Mitigating Hallucination Escape in Tool-Using LLM Agents","display_name":"Understanding and Mitigating Hallucination Escape in Tool-Using LLM Agents","publication_year":2026,"publication_date":"2026-10-03","ids":{"openalex":"https://openalex.org/W7220698552","doi":"https://doi.org/10.48550/arxiv.2610.04409"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.04409","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.04409","pdf_url":"https://arxiv.org/pdf/2610.04409","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":null,"license_id":null,"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/2610.04409","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125946081","display_name":"Peigui Qi","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Qi, Peigui","raw_affiliation_strings":["University of Science and Technology of China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5032244414","display_name":"Kunsheng Tang","orcid":"https://orcid.org/0009-0002-5690-2364","observed_orcids":["https://orcid.org/0009-0002-5690-2364"]},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tang, Kunsheng","raw_affiliation_strings":["University of Science and Technology of China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5029976990","display_name":"Yide Song","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I201448701","display_name":"University of Washington","ror":"https://ror.org/00cvxb145","country_code":"US","type":"education","lineage":["https://openalex.org/I201448701"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Song, Yide","raw_affiliation_strings":["University of Washington"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Washington","institution_ids":["https://openalex.org/I201448701"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5157677912","display_name":"Weiming Zhang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhang, Weiming","raw_affiliation_strings":["University of Science and Technology of China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China","institution_ids":["https://openalex.org/I126520041"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5157705074","display_name":"Nenghai Yu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I126520041","display_name":"University of Science and Technology of China","ror":"https://ror.org/04c4dkn09","country_code":"CN","type":"education","lineage":["https://openalex.org/I126520041","https://openalex.org/I19820366"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yu, Nenghai","raw_affiliation_strings":["University of Science and Technology of China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Science and Technology of China","institution_ids":["https://openalex.org/I126520041"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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":0.3959999978542328,"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":0.3959999978542328,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.3107999861240387,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.11299999803304672,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/tool-selection","display_name":"tool selection","score":0.843999981880188},{"id":"https://openalex.org/keywords/tool-configuration","display_name":"tool configuration","score":0.5289999842643738},{"id":"https://openalex.org/keywords/attention-enhancement","display_name":"attentional enhancement","score":0.9520000219345093},{"id":"https://openalex.org/keywords/llm-based-agents","display_name":"LLM agents","score":0.4880000054836273}],"concepts":[{"id":"https://openalex.org/C2776706361","wikidata":"https://www.wikidata.org/wiki/Q1054088","display_name":"Auditory hallucination","level":3,"score":0.571399986743927},{"id":"https://openalex.org/C2908998935","wikidata":"https://www.wikidata.org/wiki/Q130741","display_name":"Visual Hallucination","level":2,"score":0.5170999765396118},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.49219998717308044},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4253999888896942},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.4171000123023987},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.39649999141693115},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.3813000023365021},{"id":"https://openalex.org/C50335755","wikidata":"https://www.wikidata.org/wiki/Q483247","display_name":"Phenomenon","level":2,"score":0.3312999904155731},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.32710000872612},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3212999999523163},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.31279999017715454}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.04409","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.04409","pdf_url":"https://arxiv.org/pdf/2610.04409","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.04409","doi":"https://doi.org/10.48550/arxiv.2610.04409","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.04409","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.04409","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.04409","pdf_url":"https://arxiv.org/pdf/2610.04409","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7220698552.pdf","grobid_xml":"https://content.openalex.org/works/W7220698552.grobid-xml"},"referenced_works_count":5,"referenced_works":["https://openalex.org/W4389518608","https://openalex.org/W4415338087","https://openalex.org/W4416034034","https://openalex.org/W7160822010","https://openalex.org/W7172433968"],"related_works":["https://openalex.org/W7120123770","https://openalex.org/W7213675633","https://openalex.org/W7166868746","https://openalex.org/W7131898328","https://openalex.org/W4405095100","https://openalex.org/W7221811854","https://openalex.org/W4400222155","https://openalex.org/W4415428767","https://openalex.org/W7077074131","https://openalex.org/W7170076195"],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"increasingly":[4],"serve":[5],"as":[6],"autonomous":[7],"agents":[8],"that":[9,40,74,91,118],"invoke":[10],"external":[11],"tools.":[12],"However,":[13],"this":[14,70,169],"capability":[15],"introduces":[16],"tool":[17,51,88,128,177],"hallucination,":[18],"selecting":[19],"incorrect":[20],"tools":[21],"or":[22],"generating":[23],"invalid":[24],"calls.":[25],"Existing":[26],"mitigation":[27],"methods":[28,46,93],"report":[29],"substantial":[30],"improvements,":[31],"yet":[32],"we":[33,41,111],"identify":[34],"a":[35,79,114,175],"previously":[36],"overlooked":[37],"failure":[38],"mode":[39],"term":[42],"Hallucination":[43],"Escape.":[44],"These":[45],"reduce":[47],"hallucination":[48,75,105,129,132,143,149],"on":[49,56,60,108,137],"the":[50,65,86,152,181],"configuration":[52,178],"they":[53],"are":[54],"tuned":[55],"but":[57],"increase":[58],"it":[59],"other":[61],"configurations,":[62],"canceling":[63],"out":[64],"gain.":[66],"We":[67,167],"further":[68],"investigate":[69],"phenomenon":[71],"and":[72,90,147,158,179],"find":[73],"rises":[76],"sharply":[77],"when":[78],"model's":[80],"intrinsic":[81],"tool-use":[82],"tendencies":[83],"conflict":[84],"with":[85,122],"current":[87],"configuration,":[89],"existing":[92],"reinforce":[94],"rather":[95],"than":[96],"suppress":[97],"these":[98,109],"tendencies,":[99],"which":[100],"in":[101,164],"turn":[102],"contributes":[103],"to":[104,126],"escape.":[106,133],"Building":[107],"findings,":[110],"propose":[112],"EscapeGuard,":[113],"training-free":[115],"inference-time":[116],"method":[117],"combines":[119],"conflict-aware":[120],"gating":[121],"configuration-derived":[123],"attention":[124],"enhancement":[125],"mitigate":[127],"while":[130],"preventing":[131],"Across":[134],"six":[135],"benchmarks":[136],"various":[138],"models,":[139],"EscapeGuard":[140],"reduces":[141],"tool-selection":[142],"by":[144,155],"9.0":[145],"pp":[146,157],"suppresses":[148],"escape,":[150],"lowering":[151],"cross-configuration":[153],"mean":[154],"23.7":[156],"achieving":[159],"an":[160],"89.1%":[161],"net":[162],"improvement":[163],"paired-query":[165],"evaluation.":[166],"hope":[168],"work":[170],"can":[171],"encourage":[172],"evaluation":[173],"beyond":[174],"single":[176],"pave":[180],"way":[182],"for":[183],"more":[184],"reliable":[185],"tool-using":[186],"LLM":[187],"agents.":[188]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-07T00:00:00"}