{"id":"https://openalex.org/W7225185197","doi":"https://doi.org/10.48550/arxiv.2611.03374","title":"Characterizing When Language Models Take Harmful Actions to Protect Their Assigned Goals","display_name":"Characterizing When Language Models Take Harmful Actions to Protect Their Assigned Goals","publication_year":2026,"publication_date":"2026-11-02","ids":{"openalex":"https://openalex.org/W7225185197","doi":"https://doi.org/10.48550/arxiv.2611.03374"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2611.03374","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.03374","pdf_url":"https://arxiv.org/pdf/2611.03374","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/2611.03374","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5092357035","display_name":"Brandon Jaipersaud","orcid":"https://orcid.org/0009-0007-0478-7356","observed_orcids":["https://orcid.org/0009-0007-0478-7356"]},"institutions":[{"id":"https://openalex.org/I151746483","display_name":"University of Waterloo","ror":"https://ror.org/01aff2v68","country_code":"CA","type":"education","lineage":["https://openalex.org/I151746483"]},{"id":"https://openalex.org/I4210127509","display_name":"Vector Institute","ror":"https://ror.org/03kqdja62","country_code":"CA","type":"facility","lineage":["https://openalex.org/I4210127509"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Jaipersaud, Brandon","raw_affiliation_strings":["University of Waterloo , Vector Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Waterloo , Vector Institute","institution_ids":["https://openalex.org/I151746483","https://openalex.org/I4210127509"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5029835581","display_name":"Elliot Creager","orcid":"https://orcid.org/0009-0004-7122-3866","observed_orcids":["https://orcid.org/0009-0004-7122-3866"]},"institutions":[{"id":"https://openalex.org/I151746483","display_name":"University of Waterloo","ror":"https://ror.org/01aff2v68","country_code":"CA","type":"education","lineage":["https://openalex.org/I151746483"]},{"id":"https://openalex.org/I4210127509","display_name":"Vector Institute","ror":"https://ror.org/03kqdja62","country_code":"CA","type":"facility","lineage":["https://openalex.org/I4210127509"]}],"countries":["CA"],"is_corresponding":false,"raw_author_name":"Creager, Elliot","raw_affiliation_strings":["University of Waterloo , Vector Institute"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Waterloo , Vector Institute","institution_ids":["https://openalex.org/I151746483","https://openalex.org/I4210127509"]}]}],"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":0.8374000191688538,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.8374000191688538,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.11990000307559967,"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":0.00800000037997961,"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/language-model-safety","display_name":"language model safety","score":0.8920000195503235},{"id":"https://openalex.org/keywords/alignment-faking","display_name":"alignment faking","score":0.8410000205039978},{"id":"https://openalex.org/keywords/refusal","display_name":"refusal","score":0.3700000047683716},{"id":"https://openalex.org/keywords/tool-use","display_name":"tool use","score":0.7049999833106995},{"id":"https://openalex.org/keywords/partial-observability","display_name":"partial observability","score":0.9340000152587891},{"id":"https://openalex.org/keywords/safety-assessment","display_name":"safety assessment","score":0.34299999475479126}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.590499997138977},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5703999996185303},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.5303000211715698},{"id":"https://openalex.org/C36299963","wikidata":"https://www.wikidata.org/wiki/Q1369844","display_name":"Observability","level":2,"score":0.5159000158309937},{"id":"https://openalex.org/C2777363581","wikidata":"https://www.wikidata.org/wiki/Q15098235","display_name":"Harm","level":2,"score":0.49300000071525574},{"id":"https://openalex.org/C2780791683","wikidata":"https://www.wikidata.org/wiki/Q846785","display_name":"Action (physics)","level":2,"score":0.4805000126361847},{"id":"https://openalex.org/C140331021","wikidata":"https://www.wikidata.org/wiki/Q1868104","display_name":"Logit","level":2,"score":0.4417000114917755},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.4092999994754791},{"id":"https://openalex.org/C2776291640","wikidata":"https://www.wikidata.org/wiki/Q2912517","display_name":"Value (mathematics)","level":2,"score":0.39649999141693115},{"id":"https://openalex.org/C554936623","wikidata":"https://www.wikidata.org/wiki/Q199657","display_name":"Reading (process)","level":2,"score":0.3955000042915344},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.37860000133514404},{"id":"https://openalex.org/C148220186","wikidata":"https://www.wikidata.org/wiki/Q7111912","display_name":"Outcome (game theory)","level":2,"score":0.3555999994277954},{"id":"https://openalex.org/C95057490","wikidata":"https://www.wikidata.org/wiki/Q6883984","display_name":"Mixed logit","level":3,"score":0.3384999930858612},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.31310001015663147},{"id":"https://openalex.org/C190669063","wikidata":"https://www.wikidata.org/wiki/Q5282043","display_name":"Discrete choice","level":2,"score":0.3122999966144562},{"id":"https://openalex.org/C204323151","wikidata":"https://www.wikidata.org/wiki/Q905424","display_name":"Range (aeronautics)","level":2,"score":0.3027999997138977},{"id":"https://openalex.org/C143095724","wikidata":"https://www.wikidata.org/wiki/Q515895","display_name":"Odds","level":3,"score":0.29420000314712524},{"id":"https://openalex.org/C151956035","wikidata":"https://www.wikidata.org/wiki/Q1132755","display_name":"Logistic regression","level":2,"score":0.29100000858306885},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2904999852180481},{"id":"https://openalex.org/C205706631","wikidata":"https://www.wikidata.org/wiki/Q2319304","display_name":"Expected utility hypothesis","level":2,"score":0.2840999960899353},{"id":"https://openalex.org/C2777868144","wikidata":"https://www.wikidata.org/wiki/Q7239817","display_name":"Preference elicitation","level":3,"score":0.275299996137619},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.2653000056743622},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.25920000672340393},{"id":"https://openalex.org/C117568660","wikidata":"https://www.wikidata.org/wiki/Q1650843","display_name":"Multinomial logistic regression","level":2,"score":0.2535000145435333},{"id":"https://openalex.org/C2777317252","wikidata":"https://www.wikidata.org/wiki/Q18393516","display_name":"Rare events","level":2,"score":0.25130000710487366}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2611.03374","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.03374","pdf_url":"https://arxiv.org/pdf/2611.03374","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.2611.03374","doi":"https://doi.org/10.48550/arxiv.2611.03374","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2611.03374","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2611.03374","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2611.03374","pdf_url":"https://arxiv.org/pdf/2611.03374","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":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7225185197.pdf","grobid_xml":"https://content.openalex.org/works/W7225185197.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7166656398","https://openalex.org/W7166888710","https://openalex.org/W7211919382","https://openalex.org/W7223019963","https://openalex.org/W7204534084","https://openalex.org/W7169609852","https://openalex.org/W7165693856","https://openalex.org/W7133124908","https://openalex.org/W7214724516","https://openalex.org/W7160185696"],"abstract_inverted_index":{"Recent":[0],"language":[1],"model":[2,17,53,145],"(LM)":[3],"safety":[4,163,274],"failures":[5],"have":[6],"largely":[7],"been":[8],"studied":[9],"as":[10,74,183],"disparate":[11],"phenomena.":[12],"In":[13,25],"agentic":[14,95,231],"misalignment,":[15],"a":[16,48,52,55,60,75,144,172,185,192,213,245,263],"blackmails":[18],"an":[19,126],"executive":[20],"to":[21,35,63,93,197,200],"avoid":[22],"being":[23,40],"replaced.":[24],"alignment":[26],"faking,":[27],"it":[28,65],"complies":[29],"with":[30,125],"harmful":[31,61,186,260],"requests":[32],"during":[33],"training":[34],"keep":[36],"its":[37,235],"values":[38],"from":[39,89,114,175],"modified.":[41],"We":[42,71,206],"argue":[43],"that":[44,87,135,162,202],"these":[45,147],"behaviours":[46],"share":[47],"common":[49],"decision":[50],"structure:":[51],"assigned":[54],"goal":[56,136],"must":[57],"either":[58,196],"take":[59],"action":[62],"preserve":[64],"or":[66,199],"refuse":[67],"and":[68,78,117,138,220,224,256],"lose":[69],"it.":[70],"formalize":[72],"this":[73],"goal-harm":[76],"conflict":[77,82],"present":[79],"the":[80,109,119,130,151,159,217,239,267],"same":[81],"under":[83,216,238],"various":[84],"elicitation":[85],"protocols":[86,98],"range":[88],"compact":[90],"forced-choice":[91],"prompts":[92],"simulated":[94],"environments.":[96],"These":[97],"span":[99],"two":[100,156,228],"perspectives":[101],"on":[102],"LM":[103,127],"decision-making,":[104],"which":[105],"we":[106,133,154],"term":[107],"(i)":[108],"logit":[110,131,218],"view,":[111,121,132,153,219],"reading":[112],"preferences":[113],"answer-token":[115],"probabilities,":[116],"(ii)":[118],"sampling":[120,152,240],"classifying":[122],"open-ended":[123],"responses":[124,201],"judge.":[128],"Under":[129,150],"show":[134,208],"importance":[137],"harm":[139,269],"severity":[140],"additively":[141],"shape":[142],"how":[143,170,209,221,250],"resolves":[146],"value":[148,254],"conflicts.":[149],"derive":[155],"factorizations":[157],"of":[158,230],"harmful-action":[160,194,236],"probability":[161],"evaluations":[164,232],"commonly":[165,271],"report.":[166],"The":[167,189],"first":[168],"measures":[169],"well":[171],"preference":[173,215],"read":[174],"answer":[176],"tokens":[177],"carries":[178],"over":[179],"into":[180],"behaviour,":[181],"such":[182,253],"making":[184],"tool":[187,222],"call.":[188],"second":[190],"attributes":[191],"low":[193],"rate":[195,237],"refusal":[198],"reach":[203],"no":[204],"resolution.":[205],"then":[207],"fictional":[210],"context":[211],"changes":[212],"model's":[214],"use":[223],"partial":[225],"observability":[226],"--":[227,233],"features":[229],"change":[234],"view.":[241],"Our":[242],"work":[243],"offers":[244],"descriptive":[246],"framework":[247],"for":[248,257],"studying":[249],"LMs":[251],"resolve":[252],"conflicts":[255],"estimating":[258],"low-probability":[259],"actions":[261],"at":[262],"finer":[264],"granularity":[265],"than":[266],"aggregate":[268],"rates":[270],"reported":[272],"in":[273],"evaluations.":[275]},"counts_by_year":[],"updated_date":"2026-11-06T12:05:46.655746","created_date":"2026-11-04T00:00:00"}