{"id":"https://openalex.org/W7214747862","doi":"https://doi.org/10.48550/arxiv.2609.34896","title":"DeShortcut-Align: Decoupling Spurious Shortcuts for Robust Safety Alignment in Large Reasoning Models","display_name":"DeShortcut-Align: Decoupling Spurious Shortcuts for Robust Safety Alignment in Large Reasoning Models","publication_year":2026,"publication_date":"2026-09-28","ids":{"openalex":"https://openalex.org/W7214747862","doi":"https://doi.org/10.48550/arxiv.2609.34896"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.34896","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34896","pdf_url":"https://arxiv.org/pdf/2609.34896","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/2609.34896","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5153480946","display_name":"Qirui Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I76130692","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884","country_code":"CN","type":"education","lineage":["https://openalex.org/I76130692"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Liu, Qirui","raw_affiliation_strings":["The State Key Laboratory of Blockchain and Data Security , Zhejiang University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The State Key Laboratory of Blockchain and Data Security , Zhejiang University","institution_ids":["https://openalex.org/I76130692"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153488164","display_name":"Yichen Sun","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I76130692","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884","country_code":"CN","type":"education","lineage":["https://openalex.org/I76130692"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Sun, Yichen","raw_affiliation_strings":["The State Key Laboratory of Blockchain and Data Security , Zhejiang University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The State Key Laboratory of Blockchain and Data Security , Zhejiang University","institution_ids":["https://openalex.org/I76130692"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153441231","display_name":"Yan Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4405270328","display_name":"Ant Group (China)","ror":"https://ror.org/00g5w7n47","country_code":"CN","type":"company","lineage":["https://openalex.org/I4405270328"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wang, Yan","raw_affiliation_strings":["Ant Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Ant Group","institution_ids":["https://openalex.org/I4405270328"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008967163","display_name":"Zhixuan Chu","orcid":"https://orcid.org/0000-0001-6075-1816","observed_orcids":["https://orcid.org/0000-0001-6075-1816"]},"institutions":[{"id":"https://openalex.org/I76130692","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884","country_code":"CN","type":"education","lineage":["https://openalex.org/I76130692"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chu, Zhixuan","raw_affiliation_strings":["The State Key Laboratory of Blockchain and Data Security , Zhejiang University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The State Key Laboratory of Blockchain and Data Security , Zhejiang University","institution_ids":["https://openalex.org/I76130692"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088631980","display_name":"Linbo Jiang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Linbo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5077016573","display_name":"Jianan Lin","orcid":"https://orcid.org/0000-0003-4779-9609","observed_orcids":["https://orcid.org/0000-0003-4779-9609"]},"institutions":[{"id":"https://openalex.org/I4405270328","display_name":"Ant Group (China)","ror":"https://ror.org/00g5w7n47","country_code":"CN","type":"company","lineage":["https://openalex.org/I4405270328"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Lin, Jianan","raw_affiliation_strings":["Chongqing Ant Consumer Finance Co. , Ltd"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Chongqing Ant Consumer Finance Co. , Ltd","institution_ids":["https://openalex.org/I4405270328"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5121598261","display_name":"Kui Ren","orcid":"https://orcid.org/0000-0002-1969-2591","observed_orcids":["https://orcid.org/0000-0002-1969-2591","https://orcid.org/0000-0003-3441-6277"]},"institutions":[{"id":"https://openalex.org/I76130692","display_name":"Zhejiang University","ror":"https://ror.org/00a2xv884","country_code":"CN","type":"education","lineage":["https://openalex.org/I76130692"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ren, Kui","raw_affiliation_strings":["The State Key Laboratory of Blockchain and Data Security , Zhejiang University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The State Key Laboratory of Blockchain and Data Security , Zhejiang University","institution_ids":["https://openalex.org/I76130692"]}]}],"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":1,"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":0.8583999872207642,"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":0.8583999872207642,"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":0.06300000101327896,"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/T10028","display_name":"Topic Modeling","score":0.019500000402331352,"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/safety-alignment","display_name":"safety alignment","score":0.8560000061988831},{"id":"https://openalex.org/keywords/large-reasoning-models","display_name":"large reasoning models","score":0.9129999876022339},{"id":"https://openalex.org/keywords/over-refusal","display_name":"over-refusal","score":0.9559999704360962},{"id":"https://openalex.org/keywords/masked-attention","display_name":"attention masking","score":0.49399998784065247},{"id":"https://openalex.org/keywords/alignment-tax","display_name":"alignment tax","score":0.875}],"concepts":[{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.869700014591217},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7074999809265137},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.5921000242233276},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.49059998989105225},{"id":"https://openalex.org/C88006597","wikidata":"https://www.wikidata.org/wiki/Q690117","display_name":"Disk formatting","level":2,"score":0.44929999113082886},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.44429999589920044},{"id":"https://openalex.org/C108650721","wikidata":"https://www.wikidata.org/wiki/Q1783253","display_name":"Counterfactual thinking","level":2,"score":0.41990000009536743},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.40700000524520874},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.4025000035762787},{"id":"https://openalex.org/C15569618","wikidata":"https://www.wikidata.org/wiki/Q3561421","display_name":"Liveness","level":2,"score":0.3824000060558319},{"id":"https://openalex.org/C205606062","wikidata":"https://www.wikidata.org/wiki/Q5249645","display_name":"Decoupling (probability)","level":2,"score":0.3822000026702881},{"id":"https://openalex.org/C2779525943","wikidata":"https://www.wikidata.org/wiki/Q1187300","display_name":"Grammaticality","level":3,"score":0.3424000144004822},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3407999873161316},{"id":"https://openalex.org/C80156102","wikidata":"https://www.wikidata.org/wiki/Q788036","display_name":"Resolver","level":3,"score":0.3393000066280365},{"id":"https://openalex.org/C18483071","wikidata":"https://www.wikidata.org/wiki/Q168432","display_name":"Anchoring","level":2,"score":0.33649998903274536},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.32440000772476196},{"id":"https://openalex.org/C175291020","wikidata":"https://www.wikidata.org/wiki/Q1156822","display_name":"Offset (computer science)","level":2,"score":0.3203999996185303},{"id":"https://openalex.org/C122770356","wikidata":"https://www.wikidata.org/wiki/Q1656753","display_name":"Identifiability","level":2,"score":0.3052000105381012},{"id":"https://openalex.org/C134537474","wikidata":"https://www.wikidata.org/wiki/Q17144832","display_name":"Naturalness","level":2,"score":0.2851000130176544},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.2689000070095062},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2630000114440918},{"id":"https://openalex.org/C2778689934","wikidata":"https://www.wikidata.org/wiki/Q1313396","display_name":"Headline","level":2,"score":0.251800000667572}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.34896","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34896","pdf_url":"https://arxiv.org/pdf/2609.34896","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.2609.34896","doi":"https://doi.org/10.48550/arxiv.2609.34896","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.34896","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:2609.34896","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.34896","pdf_url":"https://arxiv.org/pdf/2609.34896","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/W7214747862.pdf","grobid_xml":"https://content.openalex.org/works/W7214747862.grobid-xml"},"referenced_works_count":25,"referenced_works":["https://openalex.org/W2736601468","https://openalex.org/W3177813494","https://openalex.org/W4385374425","https://openalex.org/W4385573354","https://openalex.org/W4386768656","https://openalex.org/W4391631327","https://openalex.org/W4396651342","https://openalex.org/W4399553916","https://openalex.org/W4404330874","https://openalex.org/W4405766390","https://openalex.org/W4412944723","https://openalex.org/W4414281281","https://openalex.org/W4414939355","https://openalex.org/W4414942254","https://openalex.org/W4415082370","https://openalex.org/W4415159516","https://openalex.org/W4415308126","https://openalex.org/W4416548702","https://openalex.org/W4416891281","https://openalex.org/W4417220098","https://openalex.org/W4417434884","https://openalex.org/W6948369276","https://openalex.org/W7077044503","https://openalex.org/W7083810573","https://openalex.org/W7103452980"],"related_works":["https://openalex.org/W7155451762","https://openalex.org/W4415082370","https://openalex.org/W4414981037","https://openalex.org/W7127083425","https://openalex.org/W7214938740","https://openalex.org/W4416548702","https://openalex.org/W7084387462","https://openalex.org/W4414878341","https://openalex.org/W7214592101","https://openalex.org/W7213495924"],"abstract_inverted_index":{"Safety":[0],"alignment":[1,82,107,217],"of":[2,28,50],"large":[3],"reasoning":[4,212],"models":[5,184],"(LRMs)":[6],"via":[7,164],"supervised":[8],"fine-tuning":[9],"(SFT)":[10],"and":[11,31,84,155,173,182,208],"reinforcement":[12],"learning":[13,49],"(RL)":[14],"often":[15],"yields":[16],"near-perfect":[17],"safety":[18,57,81,222],"scores,":[19],"yet":[20],"this":[21],"apparent":[22],"success":[23],"comes":[24],"at":[25],"the":[26,48,134,216],"cost":[27],"severe":[29],"over-refusal":[30,204],"degraded":[32],"general":[33],"capabilities.":[34],"Through":[35],"systematic":[36],"empirical":[37],"analysis,":[38],"we":[39,60,102],"find":[40],"that":[41,77,109,186],"these":[42,100],"failures":[43],"are":[44,70],"closely":[45],"associated":[46],"with":[47],"spurious":[51],"shortcuts":[52],"rather":[53],"than":[54],"robust":[55],"intent-sensitive":[56],"evaluation.":[58],"Specifically,":[59],"identify":[61],"two":[62],"dominant":[63],"shortcuts:":[64],"formatting":[65,176],"shortcuts,":[66,86,101],"where":[67,87],"refusal":[68,136],"behaviors":[69],"overly":[71],"bound":[72],"to":[73,129,151,167,200],"structural":[74],"prompt":[75],"templates":[76],"frequently":[78],"appear":[79],"in":[80,221],"corpora;":[83],"lexical":[85,153],"sensitive":[88],"keywords":[89],"reflexively":[90],"trigger":[91],"refusals":[92],"on":[93,99,112,133,180],"benign":[94,145],"queries.":[95],"To":[96],"mitigate":[97,152],"reliance":[98],"propose":[103],"DeShortcut-Align,":[104],"a":[105],"shortcut-decoupling":[106],"framework":[108],"reduces":[110,203],"dependence":[111],"superficial":[113],"cues.":[114],"DeShortcut-Align":[115,187],"operates":[116],"across":[117,171],"three":[118],"coordinated":[119],"stages:":[120],"(1)":[121],"Refusal":[122],"Sensitivity":[123],"Attribution,":[124],"which":[125,143,160],"masks":[126],"input":[127],"tokens":[128,150],"quantify":[130],"their":[131],"impact":[132],"final":[135],"response":[137],"distribution;":[138],"(2)":[139],"Attribution-Guided":[140],"Contrastive":[141],"Augmentation,":[142],"constructs":[144,161],"contrastive":[146],"samples":[147],"using":[148],"high-sensitivity":[149],"shortcuts;":[154],"(3)":[156],"Counterfactual":[157],"Consistency":[158],"Regularization,":[159],"template-ablated":[162],"states":[163],"attention":[165],"blinding":[166],"enforce":[168],"decision":[169],"consistency":[170],"SFT":[172],"RL,":[174],"mitigating":[175,215],"shortcut":[177],"dependence.":[178],"Experiments":[179],"7B":[181],"14B":[183],"demonstrate":[185],"significantly":[188],"improves":[189],"robustness":[190],"against":[191],"template-stripping":[192],"bypass":[193],"attacks":[194],"(reducing":[195],"performance":[196],"drops":[197],"by":[198,205],"up":[199],"72%),":[201],"substantially":[202],"over":[206],"58%,":[207],"better":[209],"preserves":[210],"general-purpose":[211],"capabilities,":[213],"thereby":[214],"tax":[218],"commonly":[219],"observed":[220],"training.":[223]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-30T00:00:00"}