{"id":"https://openalex.org/W7225734514","doi":"https://doi.org/10.54254/2755-2721/2027.38809","title":"Safety Degradation and Preservation During Benign Business Fine-Tuning","display_name":"Safety Degradation and Preservation During Benign Business Fine-Tuning","publication_year":2026,"publication_date":"2026-11-10","ids":{"openalex":"https://openalex.org/W7225734514","doi":"https://doi.org/10.54254/2755-2721/2027.38809"},"language":"en","primary_location":{"id":"doi:10.54254/2755-2721/2027.38809","doi":"https://doi.org/10.54254/2755-2721/2027.38809","is_oa":true,"landing_page_url":"https://doi.org/10.54254/2755-2721/2027.38809","pdf_url":null,"source":{"id":"https://openalex.org/S4387281889","display_name":"Applied and Computational Engineering","issn_l":"2755-2721","issn":["2755-2721","2755-273X"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Applied and Computational Engineering","raw_type":"journal-article"},"volume":null,"type":"article","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://doi.org/10.54254/2755-2721/2027.38809","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101250365","display_name":"Zekai Lin","orcid":"https://orcid.org/0009-0009-5115-9525","observed_orcids":["https://orcid.org/0009-0009-5115-9525"]},"institutions":[{"id":"https://openalex.org/I157773358","display_name":"Sun Yat-sen University","ror":"https://ror.org/0064kty71","country_code":"CN","type":"education","lineage":["https://openalex.org/I157773358"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Zekai Lin","raw_affiliation_strings":["Sun Yat-sen University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Sun Yat-sen University","institution_ids":["https://openalex.org/I157773358"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":["https://openalex.org/A5101250365"],"corresponding_institution_ids":["https://openalex.org/I157773358"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.0,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"275","issue":"1","first_page":"None","last_page":"None"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9868000149726868,"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":0.9868000149726868,"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":0.0038999998942017555,"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/T13295","display_name":"Safety Systems Engineering in Autonomy","score":0.0026000000070780516,"subfield":{"id":"https://openalex.org/subfields/2213","display_name":"Safety, Risk, Reliability and Quality"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[],"concepts":[{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.777400016784668},{"id":"https://openalex.org/C2780200862","wikidata":"https://www.wikidata.org/wiki/Q4453309","display_name":"Tweaking","level":2,"score":0.6413000226020813},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5275999903678894},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.5120999813079834},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.43650001287460327},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.3765000104904175},{"id":"https://openalex.org/C2780234205","wikidata":"https://www.wikidata.org/wiki/Q2281276","display_name":"Safety case","level":2,"score":0.37220001220703125},{"id":"https://openalex.org/C132835097","wikidata":"https://www.wikidata.org/wiki/Q7663745","display_name":"System safety","level":2,"score":0.32919999957084656},{"id":"https://openalex.org/C195094911","wikidata":"https://www.wikidata.org/wiki/Q14167904","display_name":"Process management","level":1,"score":0.28049999475479126},{"id":"https://openalex.org/C2776654903","wikidata":"https://www.wikidata.org/wiki/Q2601463","display_name":"SAFER","level":2,"score":0.2752000093460083},{"id":"https://openalex.org/C2776179734","wikidata":"https://www.wikidata.org/wiki/Q7398668","display_name":"Safety standards","level":2,"score":0.27129998803138733},{"id":"https://openalex.org/C177284502","wikidata":"https://www.wikidata.org/wiki/Q1005390","display_name":"Adapter (computing)","level":2,"score":0.2669999897480011},{"id":"https://openalex.org/C4216890","wikidata":"https://www.wikidata.org/wiki/Q815823","display_name":"Business model","level":2,"score":0.2639999985694885},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.26159998774528503},{"id":"https://openalex.org/C201995342","wikidata":"https://www.wikidata.org/wiki/Q682496","display_name":"Systems engineering","level":1,"score":0.25279998779296875},{"id":"https://openalex.org/C200601418","wikidata":"https://www.wikidata.org/wiki/Q2193887","display_name":"Reliability engineering","level":1,"score":0.2522999942302704}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.54254/2755-2721/2027.38809","doi":"https://doi.org/10.54254/2755-2721/2027.38809","is_oa":true,"landing_page_url":"https://doi.org/10.54254/2755-2721/2027.38809","pdf_url":null,"source":{"id":"https://openalex.org/S4387281889","display_name":"Applied and Computational Engineering","issn_l":"2755-2721","issn":["2755-2721","2755-273X"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Applied and Computational Engineering","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.54254/2755-2721/2027.38809","doi":"https://doi.org/10.54254/2755-2721/2027.38809","is_oa":true,"landing_page_url":"https://doi.org/10.54254/2755-2721/2027.38809","pdf_url":null,"source":{"id":"https://openalex.org/S4387281889","display_name":"Applied and Computational Engineering","issn_l":"2755-2721","issn":["2755-2721","2755-273X"],"is_oa":false,"is_in_doaj":false,"is_core":true,"listed_in":["cwts-core"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Applied and Computational Engineering","raw_type":"journal-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"While":[0],"most":[1],"Large":[2],"Language":[3],"Models":[4],"(LLMs)":[5],"are":[6],"designed":[7],"to":[8,51,165,178],"be":[9,36],"safe,":[10],"they":[11],"often":[12],"need":[13],"business":[14],"fine-tuning,":[15],"which":[16],"may":[17],"change":[18],"the":[19,26,42,53,60,80,87,97,124,127,145,159,166,181,184,186,189],"model's":[20],"inherent":[21],"safety":[22,27,54,73,113,148,195],"behaviors":[23],"and":[24,92,114,117,140,150,156,161,168,194],"weaken":[25],"guardrails":[28],"already":[29],"put":[30],"in":[31,86,123,154,192,197],"place,":[32],"something":[33],"that":[34,105],"can't":[35],"fully":[37],"managed":[38],"just":[39],"by":[40],"tweaking":[41],"training":[43,83],"data.":[44],"Thus,":[45],"it":[46,118],"becomes":[47],"an":[48],"important":[49],"problem":[50],"preserve":[52],"properties":[55],"of":[56,67,103,112,126,147,171,183,188],"a":[57,69],"model":[58,70,109],"during":[59],"fine-tuning":[61,139],"process.":[62],"This":[63],"paper":[64],"summarizes":[65],"approaches":[66],"how":[68],"can":[71],"maintain":[72],"capabilities":[74],"while":[75],"learning":[76],"new":[77],"tasks,":[78],"highlighting":[79],"interactions":[81],"between":[82],"data,":[84],"changes":[85],"internal":[88,108],"model,":[89],"protective":[90,135],"measures":[91,136],"final":[93],"evaluation.":[94],"It":[95],"reviews":[96],"research":[98,121,155,175],"literature":[99],"along":[100],"multiple":[101,198],"dimensions":[102],"factors":[104],"exacerbate":[106],"safety,":[107,193],"change,":[110],"preservation":[111],"evaluation":[115],"approaches,":[116],"highlights":[119],"specific":[120,134],"gaps":[122],"context":[125],"available":[128],"literature.":[129],"Existing":[130],"methods":[131],"have":[132,143,151],"defined":[133],"for":[137,180],"pre-,":[138],"post-training":[141],"stages,":[142],"studied":[144],"mechanisms":[146],"degradation,":[149],"made":[152],"progress":[153],"validation":[157],"from":[158],"attack":[160],"defense":[162],"sides.":[163],"Due":[164],"variety":[167],"idealized":[169],"conditions":[170],"experiments,":[172,185],"however,":[173],"additional":[174],"is":[176],"required":[177],"account":[179],"applicability":[182],"prediction":[187],"degradation":[190],"phenomena":[191],"maintenance":[196],"adapter":[199],"combinations.":[200]},"counts_by_year":[],"updated_date":"2026-11-11T12:16:26.678958","created_date":"2026-11-11T00:00:00"}