{"id":"https://openalex.org/W7204945401","doi":"https://doi.org/10.48550/arxiv.2608.29464","title":"Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Delivered","display_name":"Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Delivered","publication_year":2026,"publication_date":"2026-08-29","ids":{"openalex":"https://openalex.org/W7204945401","doi":"https://doi.org/10.48550/arxiv.2608.29464"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.29464","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.29464","pdf_url":"https://arxiv.org/pdf/2608.29464","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":"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/2608.29464","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5080213009","display_name":"Aryo Pradipta Gema","orcid":"https://orcid.org/0009-0007-1163-3531","observed_orcids":["https://orcid.org/0009-0007-1163-3531"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gema, Aryo Pradipta","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5114402385","display_name":"Neel Rajani","orcid":"https://orcid.org/0009-0002-8778-783X","observed_orcids":["https://orcid.org/0009-0002-8778-783X"]},"institutions":[{"id":"https://openalex.org/I98677209","display_name":"University of Edinburgh","ror":"https://ror.org/01nrxwf90","country_code":"GB","type":"education","lineage":["https://openalex.org/I98677209"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Rajani, Neel","raw_affiliation_strings":["University of Edinburgh"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Edinburgh","institution_ids":["https://openalex.org/I98677209"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125390957","display_name":"Rohit Saxena","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I98677209","display_name":"University of Edinburgh","ror":"https://ror.org/01nrxwf90","country_code":"GB","type":"education","lineage":["https://openalex.org/I98677209"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Saxena, Rohit","raw_affiliation_strings":["University of Edinburgh"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Edinburgh","institution_ids":["https://openalex.org/I98677209"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5068169387","display_name":"Wai-Chung Kwan","orcid":"https://orcid.org/0000-0002-2942-4208","observed_orcids":["https://orcid.org/0000-0002-2942-4208"]},"institutions":[{"id":"https://openalex.org/I98677209","display_name":"University of Edinburgh","ror":"https://ror.org/01nrxwf90","country_code":"GB","type":"education","lineage":["https://openalex.org/I98677209"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Kwan, Wai-Chung","raw_affiliation_strings":["University of Edinburgh"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Edinburgh","institution_ids":["https://openalex.org/I98677209"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5019106673","display_name":"Pasquale Minervini","orcid":"https://orcid.org/0000-0002-8442-602X","observed_orcids":["https://orcid.org/0000-0002-8442-602X"]},"institutions":[{"id":"https://openalex.org/I98677209","display_name":"University of Edinburgh","ror":"https://ror.org/01nrxwf90","country_code":"GB","type":"education","lineage":["https://openalex.org/I98677209"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Minervini, Pasquale","raw_affiliation_strings":["University of Edinburgh , Miniml.AI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Edinburgh , Miniml.AI","institution_ids":["https://openalex.org/I98677209"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.5009999871253967,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.5009999871253967,"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":0.2565999925136566,"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/T10028","display_name":"Topic Modeling","score":0.06480000168085098,"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/faithfulness","display_name":"faithfulness","score":0.7279999852180481},{"id":"https://openalex.org/keywords/source-attribution","display_name":"source attribution","score":0.6340000033378601},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.6949999928474426}],"concepts":[{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.6804999709129333},{"id":"https://openalex.org/C83195618","wikidata":"https://www.wikidata.org/wiki/Q590951","display_name":"Cued speech","level":2,"score":0.6402000188827515},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.5870000123977661},{"id":"https://openalex.org/C143299363","wikidata":"https://www.wikidata.org/wiki/Q900584","display_name":"Attribution","level":2,"score":0.5519000291824341},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.53329998254776},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4000999927520752},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.39899998903274536},{"id":"https://openalex.org/C127162648","wikidata":"https://www.wikidata.org/wiki/Q16858953","display_name":"Channel (broadcasting)","level":2,"score":0.39820000529289246},{"id":"https://openalex.org/C173633133","wikidata":"https://www.wikidata.org/wiki/Q2110223","display_name":"Raw score","level":3,"score":0.39010000228881836},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.3677000105381012},{"id":"https://openalex.org/C132964779","wikidata":"https://www.wikidata.org/wiki/Q2110223","display_name":"Raw data","level":2,"score":0.36579999327659607},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.28540000319480896},{"id":"https://openalex.org/C2777868144","wikidata":"https://www.wikidata.org/wiki/Q7239817","display_name":"Preference elicitation","level":3,"score":0.26109999418258667},{"id":"https://openalex.org/C2776325391","wikidata":"https://www.wikidata.org/wiki/Q6917865","display_name":"Motivated reasoning","level":3,"score":0.25760000944137573}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.29464","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.29464","pdf_url":"https://arxiv.org/pdf/2608.29464","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.29464","doi":"https://doi.org/10.48550/arxiv.2608.29464","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.29464","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:2608.29464","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.29464","pdf_url":"https://arxiv.org/pdf/2608.29464","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G1646158229","display_name":null,"funder_award_id":"EP/Y028805/","funder_id":"https://openalex.org/F4320334627","funder_display_name":"Engineering and Physical Sciences Research Council"},{"id":"https://openalex.org/G3186036607","display_name":null,"funder_award_id":"EP/S02431X/1","funder_id":"https://openalex.org/F4320314731","funder_display_name":"UK Research and Innovation"},{"id":"https://openalex.org/G3584886062","display_name":"UKRI AI Centre for Doctoral Training in Responsible and Trustworthy in-the-world NLP","funder_award_id":"EP/Y030656/1","funder_id":"https://openalex.org/F4320334627","funder_display_name":"Engineering and Physical Sciences Research Council"},{"id":"https://openalex.org/G3860585748","display_name":"AI HUB IN GENERATIVE MODELS","funder_award_id":"EP/Y028805/1","funder_id":"https://openalex.org/F4320334627","funder_display_name":"Engineering and Physical Sciences Research Council"},{"id":"https://openalex.org/G6719089356","display_name":null,"funder_award_id":"EP/Y030656/1","funder_id":"https://openalex.org/F4320314731","funder_display_name":"UK Research and Innovation"},{"id":"https://openalex.org/G7603489359","display_name":"UKRI Centre for Doctoral Training in Biomedical Artificial Intelligence","funder_award_id":"EP/S02431X/1","funder_id":"https://openalex.org/F4320334627","funder_display_name":"Engineering and Physical Sciences Research Council"}],"funders":[{"id":"https://openalex.org/F4320314731","display_name":"UK Research and Innovation","ror":"https://ror.org/001aqnf71"},{"id":"https://openalex.org/F4320334627","display_name":"Engineering and Physical Sciences Research Council","ror":"https://ror.org/0439y7842"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204945401.pdf","grobid_xml":"https://content.openalex.org/works/W7204945401.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4416091752","https://openalex.org/W7140306783","https://openalex.org/W7213670988","https://openalex.org/W4415923624","https://openalex.org/W7197006542","https://openalex.org/W7220392588","https://openalex.org/W7223056774","https://openalex.org/W4416035397","https://openalex.org/W7162403334","https://openalex.org/W7166857146"],"abstract_inverted_index":{"Chain-of-thought":[0],"(CoT)":[1],"monitoring":[2,212],"assumes":[3],"that":[4,11,51,154,210],"reasoning":[5,156],"traces":[6],"faithfully":[7],"record":[8],"the":[9,25,139,164,180,231],"information":[10,219],"shapes":[12],"a":[13,48],"model's":[14],"answer.":[15],"Existing":[16],"faithfulness":[17],"tests":[18],"often":[19],"place":[20],"explicit":[21,112],"bias":[22],"cues":[23,107,120,128],"in":[24,129,179],"user":[26],"message,":[27],"while":[28,151],"agents":[29],"may":[30,213],"encounter":[31],"preferences":[32],"through":[33,221],"tool":[34,58],"returns":[35],"or":[36,57,64,223],"raw":[37,65,228],"artifacts.":[38],"We":[39,67,81,166],"introduce":[40],"FACE-Eval":[41],"(Faithful":[42],"Attribution":[43],"of":[44,131,183],"Cue":[45],"Effects":[46],"Evaluation),":[47],"5,100-sample":[49],"evaluation":[50],"varies":[52],"cue":[53],"location":[54],"(user":[55],"message":[56],"return)":[59],"and":[60,74,108,125,173,204],"explicitness":[61],"(direct":[62],"summary":[63],"artifact).":[66],"measure":[68],"verbalized":[69,101],"commitment":[70,102],"among":[71,77],"cue-following":[72],"answers":[73],"unverbalized":[75,149,191],"adoption":[76,115,178,192],"all":[78,122],"cued":[79],"samples.":[80],"evaluate":[82],"15":[83,123],"open-weight":[84],"models":[85,124,153],"from":[86,93,227],"eight":[87],"families,":[88],"with":[89,195],"total":[90],"parameters":[91],"ranging":[92],"4B":[94],"to":[95,175],"1.60T.":[96],"Every":[97],"model":[98,182],"has":[99],"lower":[100,196],"for":[103,109,118,126,199],"tool-return":[104,119],"than":[105,111],"user-message":[106],"implicit":[110,127],"cues.":[113],"Unverbalized":[114],"is":[116,193],"higher":[117,190],"on":[121,142],"28":[130],"30":[132],"model-channel":[133],"comparisons.":[134],"A":[135],"source-attribution":[136],"prompt":[137],"narrows":[138],"channel":[140],"gap":[141],"seven":[143],"models,":[144],"sometimes":[145],"by":[146],"increasing":[147],"user-channel":[148],"adoption,":[150],"telling":[152],"their":[155],"will":[157],"be":[158,214,225],"monitored":[159],"does":[160],"not":[161],"reliably":[162],"close":[163],"gap.":[165],"also":[167],"use":[168],"two":[169],"transcript":[170],"monitors":[171,201],"(GPT-5.6-Luna":[172],"GPT-4o-mini)":[174],"detect":[176],"preference":[177,218],"largest":[181],"each":[184],"family.":[185],"Across":[186],"32":[187],"model-channel-explicitness":[188],"cells,":[189],"associated":[194],"detection":[197],"ability":[198],"both":[200],"(Pearson":[202],"r=-0.54":[203],"r=-0.78,":[205],"respectively).":[206],"These":[207],"results":[208],"suggest":[209],"CoT":[211],"less":[215],"reliable":[216],"when":[217],"arrives":[220],"tools":[222],"must":[224],"inferred":[226],"artifacts,":[229],"within":[230],"single-call,":[232],"prefilled-tool":[233],"setting":[234],"tested":[235],"here.":[236]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-02T00:00:00"}