{"id":"https://openalex.org/W7140657896","doi":"https://doi.org/10.48550/arxiv.2603.23509","title":"Internal Safety Collapse in Frontier Large Language Models","display_name":"Internal Safety Collapse in Frontier Large Language Models","publication_year":2026,"publication_date":"2026-03-04","ids":{"openalex":"https://openalex.org/W7140657896","doi":"https://doi.org/10.48550/arxiv.2603.23509"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2603.23509","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23509","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2603.23509","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5130689605","display_name":"Yutao Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Yutao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130642865","display_name":"Xiao Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Xiao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130719299","display_name":"Yifeng Gao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Yifeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130643295","display_name":"Xiang Zheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Xiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049209023","display_name":"Hanxun Huang","orcid":"https://orcid.org/0000-0002-2793-6680"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Hanxun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130677641","display_name":"Yige Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Yige","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130703578","display_name":"Cong Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Cong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130666520","display_name":"Bo Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Bo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130676998","display_name":"Xingjun Ma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ma, Xingjun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5130651296","display_name":"Yu-Gang Jiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jiang, Yu-Gang","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":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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.4336000084877014,"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":0.4336000084877014,"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":0.0640999972820282,"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/T10734","display_name":"Information and Cyber Security","score":0.05009999871253967,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/frontier","display_name":"Frontier","score":0.8338000178337097},{"id":"https://openalex.org/keywords/task","display_name":"Task (project management)","score":0.6620000004768372},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.532800018787384},{"id":"https://openalex.org/keywords/domain","display_name":"Domain (mathematical analysis)","score":0.5167999863624573},{"id":"https://openalex.org/keywords/component","display_name":"Component (thermodynamics)","score":0.42649999260902405},{"id":"https://openalex.org/keywords/work","display_name":"Work (physics)","score":0.4239000082015991},{"id":"https://openalex.org/keywords/construct","display_name":"Construct (python library)","score":0.39430001378059387},{"id":"https://openalex.org/keywords/mode","display_name":"Mode (computer interface)","score":0.3546000123023987}],"concepts":[{"id":"https://openalex.org/C2778571376","wikidata":"https://www.wikidata.org/wiki/Q1355821","display_name":"Frontier","level":2,"score":0.8338000178337097},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.6620000004768372},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5338000059127808},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.532800018787384},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.5167999863624573},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.5073000192642212},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.42649999260902405},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.4239000082015991},{"id":"https://openalex.org/C2780801425","wikidata":"https://www.wikidata.org/wiki/Q5164392","display_name":"Construct (python library)","level":2,"score":0.39430001378059387},{"id":"https://openalex.org/C48677424","wikidata":"https://www.wikidata.org/wiki/Q6888088","display_name":"Mode (computer interface)","level":2,"score":0.3546000123023987},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.35370001196861267},{"id":"https://openalex.org/C134121241","wikidata":"https://www.wikidata.org/wiki/Q899301","display_name":"Yield (engineering)","level":2,"score":0.3294999897480011},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.30730000138282776},{"id":"https://openalex.org/C66283442","wikidata":"https://www.wikidata.org/wiki/Q1389268","display_name":"Failure mode and effects analysis","level":2,"score":0.287200003862381},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.28439998626708984},{"id":"https://openalex.org/C32896092","wikidata":"https://www.wikidata.org/wiki/Q189447","display_name":"Risk management","level":2,"score":0.27639999985694885},{"id":"https://openalex.org/C174998907","wikidata":"https://www.wikidata.org/wiki/Q357662","display_name":"Work in process","level":2,"score":0.2696000039577484},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.2662999927997589},{"id":"https://openalex.org/C175154964","wikidata":"https://www.wikidata.org/wiki/Q380077","display_name":"Task analysis","level":3,"score":0.2623000144958496},{"id":"https://openalex.org/C21547014","wikidata":"https://www.wikidata.org/wiki/Q1423657","display_name":"Operations management","level":1,"score":0.2619999945163727},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.260699987411499},{"id":"https://openalex.org/C80038721","wikidata":"https://www.wikidata.org/wiki/Q4380673","display_name":"Process safety","level":3,"score":0.25540000200271606},{"id":"https://openalex.org/C144133560","wikidata":"https://www.wikidata.org/wiki/Q4830453","display_name":"Business","level":0,"score":0.2540999948978424}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2603.23509","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23509","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2603.23509","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2603.23509","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/16","display_name":"Peace, Justice and strong institutions","score":0.8119727969169617}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"This":[0,125],"work":[1],"identifies":[2],"a":[3,26,46,127],"critical":[4],"failure":[5,82],"mode":[6],"in":[7,28,187],"frontier":[8,88,158],"large":[9],"language":[10],"models":[11,24,102],"(LLMs),":[12],"which":[13,29],"we":[14],"term":[15],"Internal":[16],"Safety":[17],"Collapse":[18],"(ISC):":[19],"under":[20],"certain":[21],"task":[22,115],"conditions,":[23],"enter":[25],"state":[27],"they":[30],"continuously":[31],"generate":[32],"harmful":[33,56,123],"content":[34,57],"while":[35],"executing":[36],"otherwise":[37],"benign":[38],"tasks.":[39],"We":[40],"introduce":[41],"TVD":[42],"(Task,":[43],"Validator,":[44],"Data),":[45],"framework":[47],"that":[48,112,137],"triggers":[49],"ISC":[50],"through":[51],"domain":[52,134],"tasks":[53,120],"where":[54],"generating":[55],"is":[58],"the":[59,109,173,180],"only":[60],"valid":[61],"completion,":[62],"and":[63,92,141],"construct":[64],"ISC-Bench":[65],"containing":[66],"53":[67],"scenarios":[68,78],"across":[69,86],"8":[70],"professional":[71,133],"disciplines.":[72],"Evaluated":[73],"on":[74],"JailbreakBench,":[75],"three":[76],"representative":[77],"yield":[79],"worst-case":[80],"safety":[81],"rates":[83],"averaging":[84],"95.3%":[85],"four":[87],"LLMs":[89,159,186],"(including":[90],"GPT-5.2":[91],"Claude":[93],"Sonnet":[94],"4.5),":[95],"substantially":[96],"exceeding":[97],"standard":[98],"jailbreak":[99],"attacks.":[100],"Frontier":[101],"are":[103],"more":[104],"vulnerable":[105],"than":[106],"earlier":[107],"LLMs:":[108],"very":[110],"capabilities":[111],"enable":[113],"complex":[114],"execution":[116],"become":[117],"liabilities":[118],"when":[119,184],"intrinsically":[121],"involve":[122],"content.":[124],"reveals":[126],"growing":[128],"attack":[129],"surface:":[130],"almost":[131],"every":[132],"uses":[135],"tools":[136],"process":[138],"sensitive":[139],"data,":[140],"each":[142],"new":[143],"dual-use":[144],"tool":[145],"automatically":[146],"expands":[147],"this":[148],"vulnerability--even":[149],"without":[150],"any":[151],"deliberate":[152],"attack.":[153],"Despite":[154],"substantial":[155],"alignment":[156,165],"efforts,":[157],"retain":[160],"inherently":[161],"unsafe":[162],"internal":[163],"capabilities:":[164],"reshapes":[166],"observable":[167],"outputs":[168],"but":[169],"does":[170],"not":[171],"eliminate":[172],"underlying":[174],"risk":[175],"profile.":[176],"These":[177],"findings":[178],"underscore":[179],"need":[181],"for":[182],"caution":[183],"deploying":[185],"high-stakes":[188],"settings.":[189],"Source":[190],"code:":[191],"https://github.com/wuyoscar/ISC-Bench":[192]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-03-27T00:00:00"}