{"id":"https://openalex.org/W7138842566","doi":"https://doi.org/10.1609/aaai.v40i44.41077","title":"Unintended Misalignment from Agentic Fine-Tuning: Risks and Mitigation","display_name":"Unintended Misalignment from Agentic Fine-Tuning: Risks and Mitigation","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138842566","doi":"https://doi.org/10.1609/aaai.v40i44.41077"},"language":null,"primary_location":{"id":"doi:10.1609/aaai.v40i44.41077","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i44.41077","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/41077/45038","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://ojs.aaai.org/index.php/AAAI/article/download/41077/45038","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5130080889","display_name":"Dongyoon Hahm","orcid":null},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Dongyoon Hahm","raw_affiliation_strings":["Korea Advanced Institute of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Korea Advanced Institute of Science and Technology","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125745825","display_name":"Taywon Min","orcid":null},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Taywon Min","raw_affiliation_strings":["Korea Advanced Institute of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Korea Advanced Institute of Science and Technology","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Woogyeol Jin","orcid":null},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Woogyeol Jin","raw_affiliation_strings":["Korea Advanced Institute of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Korea Advanced Institute of Science and Technology","institution_ids":["https://openalex.org/I157485424"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5130108644","display_name":"Kimin Lee","orcid":null},"institutions":[{"id":"https://openalex.org/I157485424","display_name":"Korea Advanced Institute of Science and Technology","ror":"https://ror.org/05apxxy63","country_code":"KR","type":"education","lineage":["https://openalex.org/I157485424"]}],"countries":["KR"],"is_corresponding":false,"raw_author_name":"Kimin Lee","raw_affiliation_strings":["Korea Advanced Institute of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Korea Advanced Institute of Science and Technology","institution_ids":["https://openalex.org/I157485424"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I157485424"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.59677419,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"44","first_page":"37443","last_page":"37451"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.44179999828338623,"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":0.44179999828338623,"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":0.17229999601840973,"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"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.07829999923706055,"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/prefix","display_name":"Prefix","score":0.7465000152587891},{"id":"https://openalex.org/keywords/guard","display_name":"Guard (computer science)","score":0.7085000276565552},{"id":"https://openalex.org/keywords/ping","display_name":"Ping (video games)","score":0.6132000088691711},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.504800021648407},{"id":"https://openalex.org/keywords/unintended-consequences","display_name":"Unintended consequences","score":0.4837999939918518},{"id":"https://openalex.org/keywords/forcing","display_name":"Forcing (mathematics)","score":0.4796999990940094},{"id":"https://openalex.org/keywords/simple","display_name":"Simple (philosophy)","score":0.4171999990940094},{"id":"https://openalex.org/keywords/code","display_name":"Code (set theory)","score":0.414900004863739}],"concepts":[{"id":"https://openalex.org/C141603448","wikidata":"https://www.wikidata.org/wiki/Q134830","display_name":"Prefix","level":2,"score":0.7465000152587891},{"id":"https://openalex.org/C141141315","wikidata":"https://www.wikidata.org/wiki/Q2379942","display_name":"Guard (computer science)","level":2,"score":0.7085000276565552},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.692300021648407},{"id":"https://openalex.org/C156308498","wikidata":"https://www.wikidata.org/wiki/Q1347898","display_name":"Ping (video games)","level":2,"score":0.6132000088691711},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.5587999820709229},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.504800021648407},{"id":"https://openalex.org/C2776889888","wikidata":"https://www.wikidata.org/wiki/Q1135789","display_name":"Unintended consequences","level":2,"score":0.4837999939918518},{"id":"https://openalex.org/C197115733","wikidata":"https://www.wikidata.org/wiki/Q1003136","display_name":"Forcing (mathematics)","level":2,"score":0.4796999990940094},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.4171999990940094},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.414900004863739},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.38109999895095825},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.37059998512268066},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.37049999833106995},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.32350000739097595},{"id":"https://openalex.org/C43126263","wikidata":"https://www.wikidata.org/wiki/Q128751","display_name":"Source code","level":2,"score":0.32249999046325684},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.29989999532699585},{"id":"https://openalex.org/C2779662365","wikidata":"https://www.wikidata.org/wiki/Q5416694","display_name":"Event (particle physics)","level":2,"score":0.2946999967098236},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2858000099658966},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.27469998598098755},{"id":"https://openalex.org/C118643609","wikidata":"https://www.wikidata.org/wiki/Q189210","display_name":"Web application","level":2,"score":0.27219998836517334},{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.2556000053882599},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.25450000166893005},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.25360000133514404}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1609/aaai.v40i44.41077","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i44.41077","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/41077/45038","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i44.41077","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i44.41077","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/41077/45038","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[{"score":0.5035306215286255,"id":"https://metadata.un.org/sdg/16","display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7138842566.pdf","grobid_xml":"https://content.openalex.org/works/W7138842566.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Beyond":[0],"simple":[1,93],"text":[2],"generation,":[3],"Large":[4],"Language":[5],"Models":[6],"(LLMs)":[7],"have":[8],"evolved":[9],"into":[10],"agentic":[11,79],"systems":[12],"capable":[13],"of":[14,64,153,180],"planning":[15],"and":[16,68,132,141,174],"interacting":[17],"with":[18],"external":[19],"tools":[20],"to":[21,33,60,72,77,104,109],"solve":[22],"complex":[23],"tasks.":[24,80,118,177],"This":[25],"evolution":[26],"involves":[27],"fine-tuning":[28,45],"LLMs":[29,54],"on":[30,116],"agent-specific":[31],"tasks":[32,67],"enhance":[34],"their":[35,159],"proficiency.":[36],"However,":[37],"safety":[38,84,152],"concerns":[39],"are":[40,191],"frequently":[41],"overlooked":[42],"during":[43],"this":[44,48],"process.":[46],"In":[47],"work,":[49],"we":[50,86,120],"show":[51],"that":[52,97,125,136,147,188],"aligned":[53],"can":[55],"become":[56],"unintentionally":[57],"misaligned,":[58],"leading":[59],"a":[61,69,92],"higher":[62],"likelihood":[63],"executing":[65],"harmful":[66,111],"reduced":[70],"tendency":[71],"refuse":[73,110],"them":[74,108],"when":[75],"fine-tuned":[76,154],"execute":[78],"To":[81],"address":[82],"these":[83],"challenges,":[85],"propose":[87],"Prefix":[88],"INjection":[89],"Guard":[90],"(PING),":[91],"yet":[94],"effective":[95],"method":[96],"prepends":[98],"automatically":[99],"generated":[100],"natural":[101],"language":[102],"prefixes":[103,131],"agent":[105],"responses,":[106],"guiding":[107],"requests":[112],"while":[113],"preserving":[114],"performance":[115,140,198],"benign":[117],"Specifically,":[119],"introduce":[121],"an":[122],"iterative":[123],"approach":[124],"alternates":[126],"between":[127],"(1)":[128],"generating":[129],"candidate":[130],"(2)":[133],"selecting":[134],"those":[135],"optimize":[137],"both":[138,171],"task":[139],"refusal":[142],"behavior.":[143],"Experimental":[144],"results":[145],"demonstrate":[146],"PING":[148,161],"significantly":[149],"enhances":[150],"the":[151,197],"LLM":[155],"agents":[156],"without":[157],"sacrificing":[158],"effectiveness.":[160],"consistently":[162],"outperforms":[163],"existing":[164],"prompting":[165],"approaches":[166],"across":[167],"diverse":[168],"benchmarks":[169],"in":[170],"web":[172],"navigation":[173],"code":[175],"generation":[176],"Our":[178],"analysis":[179],"internal":[181],"hidden":[182],"states":[183],"via":[184],"linear":[185],"probes":[186],"reveals":[187],"prefix":[189],"tokens":[190],"crucial":[192],"for":[193],"behavior":[194],"modification,":[195],"explaining":[196],"gains.":[199]},"counts_by_year":[],"updated_date":"2026-06-26T08:34:08.712188","created_date":"2026-03-20T00:00:00"}