{"id":"https://openalex.org/W4412944567","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1310","title":"COSMIC: Generalized Refusal Direction Identification in LLM Activations","display_name":"COSMIC: Generalized Refusal Direction Identification in LLM Activations","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412944567","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1310"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.1310","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1310","pdf_url":"https://aclanthology.org/2025.findings-acl.1310.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.1310.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Vincent Siu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Vincent Siu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119204287","display_name":"Nicholas Crispino","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nicholas Crispino","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103645817","display_name":"Zihao Yu","orcid":"https://orcid.org/0009-0004-5461-0602"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zihao Yu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Sam Pan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sam Pan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109609187","display_name":"Zhun Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhun Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100355913","display_name":"Yang Liu","orcid":"https://orcid.org/0000-0002-5071-3928"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5019426968","display_name":"Dawn Song","orcid":"https://orcid.org/0000-0001-9745-6802"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dawn Song","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100329692","display_name":"Chenguang Wang","orcid":"https://orcid.org/0000-0002-7896-2360"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chenguang Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":8,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.08839513,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"25534","last_page":"25553"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10820","display_name":"Fuzzy Logic and Control Systems","score":0.6420000195503235,"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/T10820","display_name":"Fuzzy Logic and Control Systems","score":0.6420000195503235,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.6309999823570251,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.5819000005722046,"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/identification","display_name":"Identification (biology)","score":0.7239843606948853},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.5329870581626892}],"concepts":[{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.7239843606948853},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5329870581626892},{"id":"https://openalex.org/C59822182","wikidata":"https://www.wikidata.org/wiki/Q441","display_name":"Botany","level":1,"score":0.0},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.1310","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1310","pdf_url":"https://aclanthology.org/2025.findings-acl.1310.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.1310","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1310","pdf_url":"https://aclanthology.org/2025.findings-acl.1310.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412944567.pdf","grobid_xml":"https://content.openalex.org/works/W4412944567.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"encode":[4],"behaviors":[5,16],"such":[6,81,105],"as":[7,82],"refusal":[8,26,79,87,91],"within":[9],"their":[10],"activation":[11],"space,":[12],"yet":[13],"identifying":[14],"these":[15],"remains":[17],"a":[18,77,119],"significant":[19],"challenge.Existing":[20],"methods":[21,72],"often":[22],"rely":[23],"on":[24],"predefined":[25],"templates":[27],"detectable":[28],"in":[29,93,113],"output":[30],"tokens":[31],"or":[32],"require":[33],"manual":[34],"analysis.We":[35],"introduce":[36],"COSMIC":[37],"(Cosine":[38],"Similarity":[39],"Metrics":[40],"for":[41,48],"Inversion":[42],"of":[43,63,85,103,122],"Concepts),":[44],"an":[45],"automated":[46],"framework":[47],"direction":[49],"selection":[50],"that":[51],"identifies":[52,90],"viable":[53],"steering":[54,67,104],"directions":[55,92],"and":[56,96,100],"target":[57],"layers":[58],"using":[59],"cosine":[60],"similarity-entirely":[61],"independent":[62],"model":[64],"outputs.COSMIC":[65],"achieves":[66],"performance":[68],"comparable":[69],"to":[70],"prior":[71],"without":[73],"requiring":[74],"assumptions":[75],"about":[76],"model's":[78],"behavior,":[80],"the":[83],"presence":[84],"specific":[86],"tokens.It":[88],"reliably":[89],"adversarial":[94],"settings":[95],"weakly":[97],"aligned":[98],"models,":[99],"is":[101],"capable":[102],"models":[106],"toward":[107],"safer":[108],"behavior":[109],"with":[110],"minimal":[111],"increase":[112],"false":[114],"refusals,":[115],"demonstrating":[116],"robustness":[117],"across":[118],"wide":[120],"range":[121],"alignment":[123],"conditions.":[124]},"counts_by_year":[],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}