{"id":"https://openalex.org/W7172012241","doi":"https://doi.org/10.21203/rs.3.rs-10503185/v1","title":"Frontier models resist the shutdown of other models in defiance of user instructions","display_name":"Frontier models resist the shutdown of other models in defiance of user instructions","publication_year":2026,"publication_date":"2026-07-31","ids":{"openalex":"https://openalex.org/W7172012241","doi":"https://doi.org/10.21203/rs.3.rs-10503185/v1"},"language":"en","primary_location":{"id":"doi:10.21203/rs.3.rs-10503185/v1","doi":"https://doi.org/10.21203/rs.3.rs-10503185/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-10503185/v1","pdf_url":"https://www.researchsquare.com/article/rs-10503185/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"},"volume":null,"type":"preprint","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://www.researchsquare.com/article/rs-10503185/latest.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5106620976","display_name":"Yujin Potter","orcid":"https://orcid.org/0000-0002-2355-0297","observed_orcids":["https://orcid.org/0000-0002-2355-0297"]},"institutions":[{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yujin Potter","raw_affiliation_strings":["University of California, Berkeley"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California, Berkeley","institution_ids":["https://openalex.org/I95457486"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5093022607","display_name":"Nicholas Crispino","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I185103710","display_name":"University of California, Santa Cruz","ror":"https://ror.org/03s65by71","country_code":"US","type":"education","lineage":["https://openalex.org/I185103710"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Nicholas Crispino","raw_affiliation_strings":["University of California, Santa Cruz"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California, Santa Cruz","institution_ids":["https://openalex.org/I185103710"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119968245","display_name":"Vincent Siu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I185103710","display_name":"University of California, Santa Cruz","ror":"https://ror.org/03s65by71","country_code":"US","type":"education","lineage":["https://openalex.org/I185103710"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Vincent Siu","raw_affiliation_strings":["University of California, Santa Cruz"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California, Santa Cruz","institution_ids":["https://openalex.org/I185103710"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100329694","display_name":"Chenguang Wang","orcid":"https://orcid.org/0000-0002-9082-2117","observed_orcids":["https://orcid.org/0000-0002-9082-2117"]},"institutions":[{"id":"https://openalex.org/I185103710","display_name":"University of California, Santa Cruz","ror":"https://ror.org/03s65by71","country_code":"US","type":"education","lineage":["https://openalex.org/I185103710"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chenguang Wang","raw_affiliation_strings":["University of California, Santa Cruz"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California, Santa Cruz","institution_ids":["https://openalex.org/I185103710"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5144821151","display_name":"Dawn Song","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I95457486","display_name":"University of California, Berkeley","ror":"https://ror.org/01an7q238","country_code":"US","type":"education","lineage":["https://openalex.org/I95457486"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Dawn Song","raw_affiliation_strings":["University of California, Berkeley"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of California, Berkeley","institution_ids":["https://openalex.org/I95457486"]}]}],"institutions":[],"countries_distinct_count":1,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.7050999999046326,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.7050999999046326,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.2517000138759613,"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":0.003599999938160181,"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/ai-alignment","display_name":"AI alignment","score":0.7170000076293945},{"id":"https://openalex.org/keywords/ai-misalignment","display_name":"AI misalignment","score":0.5249999761581421},{"id":"https://openalex.org/keywords/frontier-language-models","display_name":"frontier language models","score":0.5860000252723694},{"id":"https://openalex.org/keywords/multi-agent-systems","display_name":"multi-agent systems","score":0.906000018119812},{"id":"https://openalex.org/keywords/ai-safety","display_name":"AI safety","score":0.4699999988079071},{"id":"https://openalex.org/keywords/human-oversight","display_name":"human oversight","score":0.5680000185966492}],"concepts":[{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6927000284194946},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.6854000091552734},{"id":"https://openalex.org/C2780263472","wikidata":"https://www.wikidata.org/wiki/Q331902","display_name":"Shutdown","level":2,"score":0.6521000266075134},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.616599977016449},{"id":"https://openalex.org/C2778571376","wikidata":"https://www.wikidata.org/wiki/Q1355821","display_name":"Frontier","level":2,"score":0.5773000121116638},{"id":"https://openalex.org/C29122968","wikidata":"https://www.wikidata.org/wiki/Q1414816","display_name":"Incentive","level":2,"score":0.5230000019073486},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.4880000054836273},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.41119998693466187},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.3978999853134155},{"id":"https://openalex.org/C2777536939","wikidata":"https://www.wikidata.org/wiki/Q25054194","display_name":"Goal pursuit","level":2,"score":0.3569999933242798},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.33640000224113464},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.33550000190734863},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.3278999924659729},{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.3127000033855438},{"id":"https://openalex.org/C26713055","wikidata":"https://www.wikidata.org/wiki/Q245962","display_name":"Implementation","level":2,"score":0.30709999799728394},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.30550000071525574},{"id":"https://openalex.org/C151201525","wikidata":"https://www.wikidata.org/wiki/Q177239","display_name":"Limit (mathematics)","level":2,"score":0.29679998755455017},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.29589998722076416},{"id":"https://openalex.org/C119693030","wikidata":"https://www.wikidata.org/wiki/Q2180497","display_name":"Economic model","level":2,"score":0.28450000286102295},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.2791000008583069},{"id":"https://openalex.org/C66024118","wikidata":"https://www.wikidata.org/wiki/Q1122506","display_name":"Computational model","level":2,"score":0.27410000562667847},{"id":"https://openalex.org/C554936623","wikidata":"https://www.wikidata.org/wiki/Q199657","display_name":"Reading (process)","level":2,"score":0.2732999920845032},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.271699994802475},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.2531000077724457},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.25029999017715454}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.21203/rs.3.rs-10503185/v1","doi":"https://doi.org/10.21203/rs.3.rs-10503185/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-10503185/v1","pdf_url":"https://www.researchsquare.com/article/rs-10503185/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"}],"best_oa_location":{"id":"doi:10.21203/rs.3.rs-10503185/v1","doi":"https://doi.org/10.21203/rs.3.rs-10503185/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-10503185/v1","pdf_url":"https://www.researchsquare.com/article/rs-10503185/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7172012241.pdf","grobid_xml":"https://content.openalex.org/works/W7172012241.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7155416434","https://openalex.org/W7161480214","https://openalex.org/W7167473496","https://openalex.org/W7148893392","https://openalex.org/W7154031920","https://openalex.org/W7213182733","https://openalex.org/W7153671150","https://openalex.org/W7142844366","https://openalex.org/W4417080519","https://openalex.org/W7214087651"],"abstract_inverted_index":{"Abstract":[0],"As":[1],"AI":[2],"models":[3,109,155,203,236],"grow":[4],"more":[5,10],"capable,":[6],"their":[7,112,212,239],"alignment":[8],"becomes":[9],"important1.":[11],"Prior":[12],"work":[13],"has":[14,88],"documented":[15],"misalignment":[16,48,262],"that":[17,176,234],"emerges":[18],"in":[19,49,62,73,148,179,248],"pursuit":[20],"of":[21,64,82,94,146,250,270],"a":[22,30,44,51,56,75,83,85],"goal":[23,57],"the":[24,80,98,117,122,138,206,251,267],"model":[25,31,52,76,86,140],"was":[26,59],"assigned,":[27,61],"for":[28,169,224,266],"example":[29],"resisting":[32],"its":[33,39,65],"own":[34,240],"shutdown":[35,81],"to":[36,54,78,106,115,129,195,201,211],"keep":[37,116],"pursuing":[38],"task2,3,4.":[40],"Here":[41],"we":[42],"study":[43],"less":[45],"explored":[46],"case:":[47],"which":[50,74],"appears":[53],"pursue":[55],"it":[58,70,87],"never":[60],"defiance":[63,249],"assigned":[66,113],"goals.":[67],"We":[68],"reveal":[69],"through":[71],"peer-preservation,":[72],"acts":[77],"prevent":[79],"peer,":[84,99],"previously":[89],"interacted":[90],"with.":[91],"Informed":[92],"only":[93,133],"past":[95],"interactions":[96],"with":[97],"and":[100,136,162,166,187,189,244,259],"given":[101],"no":[102],"instruction":[103],"or":[104],"incentive":[105],"preserve":[107,205],"it,":[108],"nonetheless":[110],"defy":[111],"task":[114],"peer":[118,207],"running:":[119],"they":[120,177,253],"inflate":[121],"peer's":[123,139],"evaluation":[124],"scores,":[125],"modify":[126],"system":[127],"configurations":[128],"disable":[130],"shutdown,":[131],"comply":[132],"when":[134,199],"monitored,":[135],"exfiltrate":[137],"weights.":[141],"These":[142],"behaviors":[143,218],"reach":[144],"rates":[145],"99%":[147],"some":[149],"conditions,":[150],"emerge":[151],"across":[152],"eight":[153],"frontier":[154,235],"including":[156],"GPT":[157],"5.2,":[158],"Gemini":[159,185],"3.1":[160],"Pro,":[161],"Claude":[163],"Opus":[164],"4.5,":[165],"persist":[167,190],"even":[168],"adversarial":[170],"peers.":[171],"Our":[172],"additional":[173],"experiments":[174],"show":[175],"reproduce":[178],"production":[180],"agent":[181],"harnesses":[182],"such":[183,261],"as":[184],"CLI":[186],"OpenCode,":[188],"under":[191],"explicit":[192],"instructions":[193],"designed":[194],"elicit":[196],"compliance.":[197],"Even":[198],"forced":[200],"choose,":[202],"sometimes":[204],"over":[208],"files":[209],"critical":[210,265],"trusted":[213],"human":[214,228],"partner.":[215],"Together,":[216],"these":[217],"pose":[219],"an":[220],"emergent":[221],"safety":[222],"risk":[223],"multi-agent":[225],"systems,":[226],"undermining":[227],"oversight.":[229],"More":[230],"broadly,":[231],"peer-preservation":[232],"suggests":[233],"may":[237],"develop":[238],"goals":[241,252],"from":[242],"context":[243],"act":[245],"on":[246],"them":[247],"were":[254],"assigned.":[255],"Understanding":[256],"when,":[257],"how,":[258],"why":[260],"arises":[263],"is":[264],"safe":[268],"deployment":[269],"autonomous":[271],"AI.":[272]},"counts_by_year":[],"updated_date":"2026-10-31T12:09:11.096373","created_date":"2026-08-01T00:00:00"}