{"id":"https://openalex.org/W7163553499","doi":"https://doi.org/10.1145/3779208.3785271","title":"Reasoning That Leaks, Fine-Tuning That Amplifies: Exposing the Hidden Threats of Chain-of-Thought Models","display_name":"Reasoning That Leaks, Fine-Tuning That Amplifies: Exposing the Hidden Threats of Chain-of-Thought Models","publication_year":2026,"publication_date":"2026-06-01","ids":{"openalex":"https://openalex.org/W7163553499","doi":"https://doi.org/10.1145/3779208.3785271"},"language":"en","primary_location":{"id":"doi:10.1145/3779208.3785271","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779208.3785271","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM Asia Conference on Computer and Communications Security","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://doi.org/10.1145/3779208.3785271","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137892787","display_name":"Zhiyuan Xu","orcid":"https://orcid.org/0009-0001-4179-1393"},"institutions":[{"id":"https://openalex.org/I36234482","display_name":"University of Bristol","ror":"https://ror.org/0524sp257","country_code":"GB","type":"education","lineage":["https://openalex.org/I36234482"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Zhiyuan Xu","raw_affiliation_strings":["School of Computer Science, University of Bristol, Bristol, United Kingdom"],"raw_orcid":"https://orcid.org/0009-0001-4179-1393","affiliations":[{"raw_affiliation_string":"School of Computer Science, University of Bristol, Bristol, United Kingdom","institution_ids":["https://openalex.org/I36234482"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102722651","display_name":"Joseph Gardiner","orcid":"https://orcid.org/0000-0003-4748-4228"},"institutions":[{"id":"https://openalex.org/I36234482","display_name":"University of Bristol","ror":"https://ror.org/0524sp257","country_code":"GB","type":"education","lineage":["https://openalex.org/I36234482"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Joseph Gardiner","raw_affiliation_strings":["School of Computer Science, University of Bristol, Bristol, United Kingdom"],"raw_orcid":"https://orcid.org/0000-0003-4748-4228","affiliations":[{"raw_affiliation_string":"School of Computer Science, University of Bristol, Bristol, United Kingdom","institution_ids":["https://openalex.org/I36234482"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5044341219","display_name":"Sana Belguith","orcid":"https://orcid.org/0000-0003-0069-8552"},"institutions":[{"id":"https://openalex.org/I36234482","display_name":"University of Bristol","ror":"https://ror.org/0524sp257","country_code":"GB","type":"education","lineage":["https://openalex.org/I36234482"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Sana Belguith","raw_affiliation_strings":["School of Computer Science, University of Bristol, Bristol, United Kingdom"],"raw_orcid":"https://orcid.org/0000-0003-0069-8552","affiliations":[{"raw_affiliation_string":"School of Computer Science, University of Bristol, Bristol, United Kingdom","institution_ids":["https://openalex.org/I36234482"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I36234482"],"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":"439","last_page":"455"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11883","display_name":"Embodied and Extended Cognition","score":0.1152999997138977,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},"topics":[{"id":"https://openalex.org/T11883","display_name":"Embodied and Extended Cognition","score":0.1152999997138977,"subfield":{"id":"https://openalex.org/subfields/2805","display_name":"Cognitive Neuroscience"},"field":{"id":"https://openalex.org/fields/28","display_name":"Neuroscience"},"domain":{"id":"https://openalex.org/domains/1","display_name":"Life Sciences"}},{"id":"https://openalex.org/T10258","display_name":"Philosophy and Theoretical Science","score":0.07209999859333038,"subfield":{"id":"https://openalex.org/subfields/3205","display_name":"Experimental and Cognitive Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11010","display_name":"Logic, Reasoning, and Knowledge","score":0.05420000106096268,"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/set","display_name":"Set (abstract data type)","score":0.2754000127315521},{"id":"https://openalex.org/keywords/perspective","display_name":"Perspective (graphical)","score":0.2709999978542328},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.2703000009059906},{"id":"https://openalex.org/keywords/identification","display_name":"Identification (biology)","score":0.26660001277923584},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.25130000710487366}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.535099983215332},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.33180001378059387},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3310000002384186},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.2754000127315521},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.2709999978542328},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2703000009059906},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.26660001277923584},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.25130000710487366},{"id":"https://openalex.org/C9652623","wikidata":"https://www.wikidata.org/wiki/Q190109","display_name":"Field (mathematics)","level":2,"score":0.2484000027179718},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.23639999330043793}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1145/3779208.3785271","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779208.3785271","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM Asia Conference on Computer and Communications Security","raw_type":"proceedings-article"},{"id":"pmh:oai:research-information.bris.ac.uk:openaire/df4bad6f-1452-4711-8070-f831064a4425","is_oa":true,"landing_page_url":"https://research-information.bris.ac.uk/en/publications/df4bad6f-1452-4711-8070-f831064a4425","pdf_url":null,"source":{"id":"https://openalex.org/S7407055359","display_name":"Explore Bristol Research","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Xu, Z, Gardiner, J & Belguith, S 2026, Reasoning That Leaks, Fine-Tuning That Amplifies : Exposing the Hidden Threats of Chain-of-Thought Models. in ASIA CCS 2026 - Proceedings of the 21st ACM ASIA Conference on Computer and Communications Security : Proceedings of the ACM Asia Conference on Computer and Communications Security. Proceedings of the ACM Asia Conference on Computer and Communications Security, Association for Computing Machinery, pp. 439-455. https://doi.org/10.1145/3779208.3785271","raw_type":"info:eu-repo/semantics/conferenceObject"}],"best_oa_location":{"id":"doi:10.1145/3779208.3785271","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3779208.3785271","pdf_url":null,"source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the ACM Asia Conference on Computer and Communications Security","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":11,"referenced_works":["https://openalex.org/W2973125940","https://openalex.org/W3100806282","https://openalex.org/W3159959439","https://openalex.org/W4205991051","https://openalex.org/W4306845373","https://openalex.org/W4396882206","https://openalex.org/W4402264526","https://openalex.org/W4402683797","https://openalex.org/W4405181291","https://openalex.org/W4411113009","https://openalex.org/W7133233472"],"related_works":[],"abstract_inverted_index":{"Chain-of-Thought":[0],"(CoT)":[1],"guides":[2],"large":[3],"language":[4],"models":[5,72,83],"to":[6,58],"reason":[7],"step-by-step,":[8],"yielding":[9],"remarkable":[10],"performance":[11],"gains":[12],"across":[13],"diverse":[14],"tasks.":[15],"However,":[16],"this":[17,29],"structured":[18],"reasoning":[19,94,123,188],"process":[20,95,124,189],"also":[21],"introduces":[22],"novel":[23],"and":[24,73,100,125,137,164,180,185],"underexplored":[25],"security":[26],"risks.":[27],"In":[28],"paper,":[30],"we":[31,77,129,153],"present":[32],"an":[33,155],"in-depth":[34],"analysis":[35],"of":[36,145],"fine-tuning":[37],"attacks":[38],"targeting":[39],"CoT-enabled":[40],"LLMs,":[41],"with":[42],"particular":[43],"focus":[44],"on":[45,69],"\u201caha":[46],"moments\u201d":[47],"during":[48],"reasoning,":[49],"which":[50],"are":[51],"critical":[52],"intermediate":[53],"steps":[54],"the":[55,93,104,109,118,122,126,143,160,187],"model":[56],"takes":[57],"make":[59],"a":[60,113,166],"significant":[61],"decision":[62],"or":[63],"change":[64],"its":[65],"behavior.":[66],"Through":[67],"experiments":[68],"six":[70],"CoT":[71,82,171,179],"three":[74],"non-CoT":[75],"baselines,":[76],"find":[78],"that":[79,140],"even":[80,107],"aligned":[81],"can":[84],"be":[85],"more":[86,98],"harmful":[87,99,114,146],"than":[88,103],"their":[89],"base":[90],"models.":[91],"Moreover,":[92],"frequently":[96],"contains":[97],"actionable":[101],"content":[102],"final":[105,110,127],"answer,":[106],"when":[108],"answer":[111],"refuses":[112],"request.":[115],"By":[116],"examining":[117],"causal":[119],"relationship":[120],"between":[121],"outputs,":[128],"identify":[130],"two":[131],"distinct":[132],"failure":[133],"modes,":[134],"Unintentional":[135],"Leakage":[136],"Harmful":[138],"Escalation,":[139],"systematically":[141],"drive":[142],"generation":[144],"reasoning.":[147,172],"To":[148],"rigorously":[149],"assess":[150],"these":[151],"risks,":[152],"propose":[154],"evaluation":[156],"framework":[157],"grounded":[158],"in":[159,178,190],"EU":[161],"AI":[162],"Act":[163],"construct":[165],"policy-aligned":[167],"benchmark":[168],"dataset":[169],"for":[170,183],"Our":[173],"findings":[174],"expose":[175],"inherent":[176],"vulnerabilities":[177],"offer":[181],"insights":[182],"supervising":[184],"aligning":[186],"LLMs.":[191]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2026-06-05T00:00:00"}