{"id":"https://openalex.org/W7166890832","doi":"https://doi.org/10.18653/v1/2026.acl-long.1616","title":"Can LLM Safety Be Ensured by Constraining Parameter Regions?","display_name":"Can LLM Safety Be Ensured by Constraining Parameter Regions?","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166890832","doi":"https://doi.org/10.18653/v1/2026.acl-long.1616"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-long.1616","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1616","pdf_url":"https://aclanthology.org/2026.acl-long.1616.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.acl-long.1616.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5126603494","display_name":"Zongmin Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zongmin Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139803121","display_name":"Jian Su","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jian Su","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075313163","display_name":"Farah Benamara","orcid":"https://orcid.org/0000-0002-0685-1864"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Farah Benamara","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5126606197","display_name":"Aixin Sun","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Aixin Sun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"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":0.85692828,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"34979","last_page":"35011"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10142","display_name":"Formal Methods in Verification","score":0.1014999970793724,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T10142","display_name":"Formal Methods in Verification","score":0.1014999970793724,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/T10126","display_name":"Logic, programming, and type systems","score":0.04910000041127205,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.04639999940991402,"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/work","display_name":"Work (physics)","score":0.2847000062465668},{"id":"https://openalex.org/keywords/troubleshooting","display_name":"Troubleshooting","score":0.26010000705718994},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.258899986743927},{"id":"https://openalex.org/keywords/stability","display_name":"Stability (learning theory)","score":0.25279998779296875},{"id":"https://openalex.org/keywords/reliability","display_name":"Reliability (semiconductor)","score":0.2313999980688095}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3659999966621399},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.33500000834465027},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2847000062465668},{"id":"https://openalex.org/C39432304","wikidata":"https://www.wikidata.org/wiki/Q188847","display_name":"Environmental science","level":0,"score":0.2721000015735626},{"id":"https://openalex.org/C147494362","wikidata":"https://www.wikidata.org/wiki/Q2078905","display_name":"Troubleshooting","level":2,"score":0.26010000705718994},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.258899986743927},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.25279998779296875},{"id":"https://openalex.org/C43214815","wikidata":"https://www.wikidata.org/wiki/Q7310987","display_name":"Reliability (semiconductor)","level":3,"score":0.2313999980688095},{"id":"https://openalex.org/C17500928","wikidata":"https://www.wikidata.org/wiki/Q959968","display_name":"Control system","level":2,"score":0.22460000216960907},{"id":"https://openalex.org/C2778348673","wikidata":"https://www.wikidata.org/wiki/Q739302","display_name":"Production (economics)","level":2,"score":0.22190000116825104}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-long.1616","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1616","pdf_url":"https://aclanthology.org/2026.acl-long.1616.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.acl-long.1616","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1616","pdf_url":"https://aclanthology.org/2026.acl-long.1616.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/13","display_name":"Climate action","score":0.6893115639686584}],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166890832.pdf","grobid_xml":"https://content.openalex.org/works/W7166890832.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"are":[4,77],"often":[5],"assumed":[6],"to":[7,36,63,92,112,121,123],"contain":[8],"\"safety":[9],"regions\"":[10],"-parameter":[11],"subsets":[12],"whose":[13],"modification":[14],"directly":[15],"influences":[16],"safety":[17,25,50,58,75,98,143],"behaviors.We":[18],"conduct":[19],"a":[20,95],"systematic":[21],"evaluation":[22],"of":[23,43,148],"four":[24,41,106],"region":[26],"identification":[27,51],"methods":[28,107],"spanning":[29],"different":[30],"parameter":[31],"granularities,":[32],"from":[33],"individual":[34],"weights":[35],"entire":[37],"Transformer":[38],"layers,":[39],"across":[40],"families":[42],"backbone":[44],"LLMs":[45],"with":[46],"varying":[47],"sizes.Using":[48],"ten":[49],"datasets,":[52],"we":[53],"find":[54],"that":[55,88],"the":[56,74,105,124,146],"identified":[57],"regions":[59,76,144],"exhibit":[60],"only":[61],"low":[62],"moderate":[64],"overlap,":[65],"as":[66],"measured":[67],"by":[68],"IoU.The":[69],"overlap":[70],"drops":[71],"significantly":[72],"when":[73],"further":[78],"refined":[79],"using":[80],"utility":[81],"datasets":[82],"(i.e.,":[83],"non-harmful":[84],"queries).These":[85],"results":[86],"suggest":[87],"current":[89],"techniques":[90],"fail":[91],"reliably":[93],"identify":[94,142],"stable,":[96],"dataset-agnostic":[97],"region.":[99],"1":[100],"PreliminariesWe":[101],"now":[102],"briefly":[103],"review":[104],"examined":[108],"in":[109],"this":[110],"study":[111],"provide":[113],"background":[114],"for":[115,127],"our":[116],"empirical":[117],"analysis.We":[118],"encourage":[119],"readers":[120],"refer":[122],"original":[125],"papers":[126],"full":[128],"technical":[129],"details.":[130],"SNIP":[131],"&":[132,137],"Wanda,":[133,138],"SafeNeuron":[134],"and":[135,140],"NLSRSNIP":[136],"SafeNeuron,":[139],"NLSR":[141],"at":[145],"levels":[147],"parameters,":[149],"neurons,":[150]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}