{"id":"https://openalex.org/W4416615468","doi":"https://doi.org/10.48550/arxiv.2510.19851","title":"Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability","display_name":"Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability","publication_year":2025,"publication_date":"2025-10-21","ids":{"openalex":"https://openalex.org/W4416615468","doi":"https://doi.org/10.48550/arxiv.2510.19851"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2510.19851","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2510.19851","pdf_url":"https://arxiv.org/pdf/2510.19851","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2510.19851","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137796756","display_name":"Artur Zolkowski","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zolkowski, Artur","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Xing, Wen","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xing, Wen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5059597480","display_name":"David Lindner","orcid":"https://orcid.org/0000-0001-7051-7433","observed_orcids":["https://orcid.org/0000-0001-7051-7433"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lindner, David","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5006851333","display_name":"Florian Tram\u00e8r","orcid":"https://orcid.org/0000-0001-8703-8762","observed_orcids":["https://orcid.org/0000-0001-8703-8762"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tram\u00e8r, Florian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5029945192","display_name":"Erik Jenner","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jenner, Erik","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.5562000274658203,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.5562000274658203,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.25679999589920044,"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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.1469999998807907,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/chain-of-thought","display_name":"chain-of-thought","score":0.7879999876022339},{"id":"https://openalex.org/keywords/alignment-monitoring","display_name":"alignment monitoring","score":0.8059999942779541},{"id":"https://openalex.org/keywords/deceptive-alignment","display_name":"deceptive alignment","score":0.7139999866485596},{"id":"https://openalex.org/keywords/model-monitoring","display_name":"model monitoring","score":0.3709999918937683},{"id":"https://openalex.org/keywords/prompt-engineering","display_name":"prompt engineering","score":0.5569999814033508},{"id":"https://openalex.org/keywords/adversarial-prompts","display_name":"adversarial prompts","score":0.7350000143051147},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7350000143051147}],"concepts":[{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.8367000222206116},{"id":"https://openalex.org/C40305131","wikidata":"https://www.wikidata.org/wiki/Q2616305","display_name":"Obfuscation","level":2,"score":0.7427999973297119},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7159000039100647},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5293999910354614},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5009999871253967},{"id":"https://openalex.org/C58642233","wikidata":"https://www.wikidata.org/wiki/Q8269924","display_name":"Taxonomy (biology)","level":2,"score":0.3928999900817871},{"id":"https://openalex.org/C41065033","wikidata":"https://www.wikidata.org/wiki/Q2825412","display_name":"Adversary","level":2,"score":0.3668000102043152},{"id":"https://openalex.org/C182306322","wikidata":"https://www.wikidata.org/wiki/Q1779371","display_name":"Order (exchange)","level":2,"score":0.34769999980926514},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.33320000767707825},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.3244999945163727},{"id":"https://openalex.org/C2780589192","wikidata":"https://www.wikidata.org/wiki/Q7285140","display_name":"Raising (metalworking)","level":2,"score":0.2883000075817108},{"id":"https://openalex.org/C66024118","wikidata":"https://www.wikidata.org/wiki/Q1122506","display_name":"Computational model","level":2,"score":0.27810001373291016},{"id":"https://openalex.org/C86827895","wikidata":"https://www.wikidata.org/wiki/Q7098582","display_name":"Opportunistic reasoning","level":4,"score":0.2542000114917755}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2510.19851","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2510.19851","pdf_url":"https://arxiv.org/pdf/2510.19851","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.2510.19851","doi":"https://doi.org/10.48550/arxiv.2510.19851","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2510.19851","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:2510.19851","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2510.19851","pdf_url":"https://arxiv.org/pdf/2510.19851","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":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7161587834","https://openalex.org/W7127201241","https://openalex.org/W4415971848","https://openalex.org/W7214625344","https://openalex.org/W7214890021","https://openalex.org/W4416716728","https://openalex.org/W7131071232","https://openalex.org/W4416435807","https://openalex.org/W4415923624","https://openalex.org/W7172513225"],"abstract_inverted_index":{"Recent":[0],"findings":[1],"suggest":[2,152],"that":[3,153],"misaligned":[4],"models":[5,25,47,125],"may":[6],"exhibit":[7],"deceptive":[8],"behavior,":[9],"raising":[10],"concerns":[11],"about":[12],"output":[13],"trustworthiness.":[14],"Chain-of-thought":[15],"(CoT)":[16],"is":[17],"a":[18,42,72],"promising":[19],"tool":[20],"for":[21],"alignment":[22],"monitoring:":[23],"when":[24],"articulate":[26],"their":[27,49,138,144],"reasoning":[28,94],"faithfully,":[29],"monitors":[30],"can":[31],"detect":[32],"and":[33,65,74,90,100,114],"mitigate":[34],"harmful":[35],"behaviors":[36],"before":[37],"undesirable":[38],"outcomes":[39],"occur.":[40],"However,":[41],"key":[43],"uncertainty":[44],"is:":[45],"Can":[46],"obfuscate":[48,137],"CoT":[50,68,81,87,92,110,140,146,155],"in":[51,95,104,159],"order":[52],"to":[53,79],"pursue":[54],"hidden":[55],"adversarial":[56,128],"objectives":[57],"while":[58,130,154],"evading":[59,131],"detection?":[60],"To":[61],"answer":[62],"this":[63],"question":[64],"thus":[66],"stress-test":[67],"monitorability,":[69],"we":[70],"develop":[71],"composable":[73],"quantifiable":[75],"taxonomy":[76],"of":[77,167],"prompts":[78],"elicit":[80],"obfuscation.":[82],"We":[83,106],"evaluate":[84],"both":[85],"internal":[86,139],"(reasoning":[88],"traces)":[89],"external":[91,145],"(prompted":[93],"outputs)":[96],"using":[97],"toy":[98],"tasks":[99,129],"more":[101],"realistic":[102],"environments":[103],"SHADE-Arena.":[105],"show":[107],"that:":[108],"(i)":[109],"monitoring":[111],"performs":[112],"accurately":[113],"efficiently":[115],"without":[116],"obfuscation":[117,122],"pressure.":[118],"(ii)":[119],"Under":[120],"strong":[121],"pressure,":[123],"some":[124],"successfully":[126],"complete":[127],"detection.":[132],"(iii)":[133],"Models":[134],"do":[135],"not":[136],"as":[141,143],"much":[142],"(under":[147],"prompt":[148],"pressure).":[149],"These":[150],"results":[151],"provides":[156],"valuable":[157],"oversight":[158],"benign":[160],"settings,":[161],"robust":[162],"deployment":[163],"requires":[164],"model-specific":[165],"stress-testing":[166],"monitorability.":[168]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2025-10-25T00:00:00"}