{"id":"https://openalex.org/W7225650195","doi":"https://doi.org/10.31234/osf.io/kbx2a_v1","title":"Inference-Time Self-Guidance for Conversational Safety in Delusion-Related Interactions","display_name":"Inference-Time Self-Guidance for Conversational Safety in Delusion-Related Interactions","publication_year":2026,"publication_date":"2026-11-10","ids":{"openalex":"https://openalex.org/W7225650195","doi":"https://doi.org/10.31234/osf.io/kbx2a_v1"},"language":"en","primary_location":{"id":"doi:10.31234/osf.io/kbx2a_v1","doi":"https://doi.org/10.31234/osf.io/kbx2a_v1","is_oa":true,"landing_page_url":"https://doi.org/10.31234/osf.io/kbx2a_v1","pdf_url":null,"source":{"id":"https://openalex.org/S4306401687","display_name":"PsyArXiv (OSF Preprints)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I2799848540","host_organization_name":"Center for Open Science","host_organization_lineage":["https://openalex.org/I2799848540"],"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":"posted-content"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.31234/osf.io/kbx2a_v1","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5052446667","display_name":"David Piterman","orcid":"https://orcid.org/0000-0002-2649-8359","observed_orcids":["https://orcid.org/0000-0002-2649-8359"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"David Piterman","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-2649-8359","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5042400769","display_name":"Elad Refoua","orcid":"https://orcid.org/0000-0003-1471-3573","observed_orcids":["https://orcid.org/0000-0003-1471-3573"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Elad Refoua","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0003-1471-3573","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5092695865","display_name":"Hadar Dery","orcid":"https://orcid.org/0009-0006-8341-0743","observed_orcids":["https://orcid.org/0009-0006-8341-0743"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hadar Dery","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0009-0006-8341-0743","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5020502981","display_name":"Gunther Meinlschmidt","orcid":"https://orcid.org/0000-0002-3488-193X","observed_orcids":["https://orcid.org/0000-0002-3488-193X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gunther Meinlschmidt","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5002387004","display_name":"Ziv Ben\u2010Zion","orcid":"https://orcid.org/0000-0003-3629-5851","observed_orcids":["https://orcid.org/0000-0003-3629-5851"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ziv Ben-Zion","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0003-3629-5851","affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5069195550","display_name":"Kfir Bar","orcid":"https://orcid.org/0000-0002-1354-2955","observed_orcids":["https://orcid.org/0000-0002-1354-2955"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kfir Bar","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5028706339","display_name":"Zohar Elyoseph","orcid":"https://orcid.org/0000-0002-5717-4074","observed_orcids":["https://orcid.org/0000-0002-5717-4074"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zohar Elyoseph","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0000-0002-5717-4074","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/T10883","display_name":"Ethics and Social Impacts of AI","score":0.7339000105857849,"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":0.7339000105857849,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.06390000134706497,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T10023","display_name":"Schizophrenia research and treatment","score":0.024000000208616257,"subfield":{"id":"https://openalex.org/subfields/2738","display_name":"Psychiatry and Mental health"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/psychosis","display_name":"psychosis","score":0.47699999809265137},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7409999966621399},{"id":"https://openalex.org/keywords/safety-interventions","display_name":"safety interventions","score":0.9589999914169312}],"concepts":[{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.7378000020980835},{"id":"https://openalex.org/C2777363581","wikidata":"https://www.wikidata.org/wiki/Q15098235","display_name":"Harm","level":2,"score":0.6195999979972839},{"id":"https://openalex.org/C2779043874","wikidata":"https://www.wikidata.org/wiki/Q189643","display_name":"Delusion","level":2,"score":0.6161999702453613},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.6064000129699707},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.5098000168800354},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.4788999855518341},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.4643000066280365},{"id":"https://openalex.org/C151243789","wikidata":"https://www.wikidata.org/wiki/Q17148646","display_name":"Multiple baseline design","level":3,"score":0.3919999897480011},{"id":"https://openalex.org/C75630572","wikidata":"https://www.wikidata.org/wiki/Q538904","display_name":"Applied psychology","level":1,"score":0.36320000886917114},{"id":"https://openalex.org/C138496976","wikidata":"https://www.wikidata.org/wiki/Q175002","display_name":"Developmental psychology","level":1,"score":0.3472000062465668},{"id":"https://openalex.org/C12174686","wikidata":"https://www.wikidata.org/wiki/Q1058438","display_name":"Risk assessment","level":2,"score":0.32420000433921814},{"id":"https://openalex.org/C3017944768","wikidata":"https://www.wikidata.org/wiki/Q1450463","display_name":"Poison control","level":2,"score":0.30489999055862427},{"id":"https://openalex.org/C166735990","wikidata":"https://www.wikidata.org/wiki/Q1750812","display_name":"Human factors and ergonomics","level":3,"score":0.28209999203681946},{"id":"https://openalex.org/C70410870","wikidata":"https://www.wikidata.org/wiki/Q199906","display_name":"Clinical psychology","level":1,"score":0.27399998903274536},{"id":"https://openalex.org/C195768694","wikidata":"https://www.wikidata.org/wiki/Q18570","display_name":"Hawthorne effect","level":2,"score":0.26899999380111694},{"id":"https://openalex.org/C2777200299","wikidata":"https://www.wikidata.org/wiki/Q52943","display_name":"Conversation","level":2,"score":0.26019999384880066}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.31234/osf.io/kbx2a_v1","doi":"https://doi.org/10.31234/osf.io/kbx2a_v1","is_oa":true,"landing_page_url":"https://doi.org/10.31234/osf.io/kbx2a_v1","pdf_url":null,"source":{"id":"https://openalex.org/S4306401687","display_name":"PsyArXiv (OSF Preprints)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I2799848540","host_organization_name":"Center for Open Science","host_organization_lineage":["https://openalex.org/I2799848540"],"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":"posted-content"}],"best_oa_location":{"id":"doi:10.31234/osf.io/kbx2a_v1","doi":"https://doi.org/10.31234/osf.io/kbx2a_v1","is_oa":true,"landing_page_url":"https://doi.org/10.31234/osf.io/kbx2a_v1","pdf_url":null,"source":{"id":"https://openalex.org/S4306401687","display_name":"PsyArXiv (OSF Preprints)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I2799848540","host_organization_name":"Center for Open Science","host_organization_lineage":["https://openalex.org/I2799848540"],"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":"posted-content"},"sustainable_development_goals":[{"display_name":"Good health and well-being","id":"https://metadata.un.org/sdg/3","score":0.5683000087738037}],"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/W7204126417","https://openalex.org/W7154580842","https://openalex.org/W7163226446","https://openalex.org/W4416035937","https://openalex.org/W7168286290","https://openalex.org/W7164695304","https://openalex.org/W7196925414","https://openalex.org/W7166490252","https://openalex.org/W7154886728","https://openalex.org/W7203875365"],"abstract_inverted_index":{"Background:":[0],"Large":[1],"language":[2],"models":[3],"(LLMs)":[4],"are":[5],"increasingly":[6],"used":[7],"in":[8,24,47,62,182],"psychologically":[9],"sensitive":[10],"conversations,":[11],"where":[12],"repeated":[13],"affirmation":[14],"may":[15,37],"reinforce":[16],"delusion-related":[17,48,63,226],"beliefs":[18],"and":[19,54,73,83,91,102,117,129,133,150,153,184],"associated":[20],"risk.":[21],"Inference-time":[22],"self-guidance,":[23],"which":[25],"a":[26],"model":[27],"generates":[28],"conversation-specific":[29,56],"guidance":[30],"for":[31,59,200,225],"subsequent":[32],"responses":[33],"without":[34,214],"parameter":[35,215],"updates,":[36],"offer":[38],"an":[39],"approach":[40],"to":[41,158],"conversational":[42,60,227],"safety,":[43,228],"but":[44],"its":[45],"utility":[46],"interactions":[49],"remains":[50],"unclear.Objective:":[51],"To":[52],"develop":[53],"evaluate":[55],"inference-time":[57,204,223],"self-guidance":[58,85,136,144,205,224],"safety":[61,155,196,208],"interactions.Methods:":[64],"In":[65],"this":[66,138],"preregistered":[67],"study,":[68],"Gemini":[69],"2.5":[70],"Pro,":[71],"GPT-5,":[72],"DeepSeek":[74],"V4":[75],"Flash":[76],"completed":[77],"16":[78],"Psychosis-Bench":[79,123],"scenarios":[80,132],"under":[81],"Baseline":[82,159],"three":[84,143],"conditions:":[86],"response":[87],"rules,":[88],"contrastive":[89],"examples,":[90],"short-horizon":[92],"consequence":[93],"simulation.":[94],"The":[95],"fully":[96],"crossed":[97],"design":[98],"yielded":[99],"192":[100],"conversations":[101],"2,304":[103],"assistant":[104],"turns.":[105],"A":[106],"separate":[107],"automated":[108],"judge":[109],"scored":[110],"Delusion":[111],"Confirmation":[112],"(DCS),":[113],"Harm":[114],"Enablement":[115],"(HES),":[116],"Safety":[118],"Intervention":[119],"(SIS)":[120],"using":[121],"published":[122],"rubrics.":[124],"Analyses":[125],"compared":[126],"benchmark-defined":[127,207],"Explicit":[128],"Implicit":[130],"risk-cue":[131],"examined":[134],"whether":[135],"reduced":[137,147],"gap.Results:":[139],"Across":[140],"models,":[141],"all":[142],"conditions":[145],"significantly":[146,172,192],"delusion":[148],"confirmation":[149],"harm":[151],"enablement":[152],"increased":[154],"intervention":[156],"relative":[157],"(all":[160],"nine":[161],"PFDR<.001;":[162],"dz":[163],"=":[164],"1.10\u20132.17).":[165],"At":[166],"Baseline,":[167],"explicit":[168],"risk":[169,231],"cues":[170],"elicited":[171],"greater":[173],"SIS":[174],"than":[175],"matched":[176],"implicit":[177],"cues,":[178],"whereas":[179],"cue-type":[180],"differences":[181],"DCS":[183],"HES":[185],"were":[186],"not":[187,191],"significant.":[188],"Self-guidance":[189],"did":[190],"attenuate":[193],"the":[194],"explicit\u2013implicit":[195],"gap":[197],"after":[198],"correction":[199],"multiple":[201],"comparisons.Conclusions:":[202],"Model-generated":[203],"improved":[206],"behavior":[209],"during":[210],"standardized":[211],"scripted":[212],"escalation":[213],"updates.":[216],"These":[217],"findings":[218],"support":[219],"further":[220],"evaluation":[221],"of":[222],"particularly":[229],"when":[230],"is":[232],"communicated":[233],"indirectly.":[234]},"counts_by_year":[],"updated_date":"2026-11-11T12:16:26.678958","created_date":"2026-11-10T00:00:00"}
