{"id":"https://openalex.org/W4411713638","doi":"https://doi.org/10.1145/3716368.3735302","title":"On Jailbreaking Quantized Language Models Through Fault Injection Attacks","display_name":"On Jailbreaking Quantized Language Models Through Fault Injection Attacks","publication_year":2025,"publication_date":"2025-06-27","ids":{"openalex":"https://openalex.org/W4411713638","doi":"https://doi.org/10.1145/3716368.3735302"},"language":"en","primary_location":{"id":"doi:10.1145/3716368.3735302","doi":"https://doi.org/10.1145/3716368.3735302","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3716368.3735302","pdf_url":null,"source":{"id":"https://openalex.org/S4363608736","display_name":"ACM Great Lakes Symposium on VLSI (GLSVLSI)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the Great Lakes Symposium on VLSI 2025","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://doi.org/10.1145/3716368.3735302","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5114414752","display_name":"Noureldin Zahran","orcid":"https://orcid.org/0009-0006-4471-5026","observed_orcids":["https://orcid.org/0009-0006-4471-5026"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Noureldin Zahran","raw_affiliation_strings":["Compumacy for Artificial Intelligence solutions, Cairo, Egypt"],"raw_orcid":"https://orcid.org/0009-0006-4471-5026","affiliations":[{"raw_affiliation_string":"Compumacy for Artificial Intelligence solutions, Cairo, Egypt","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5118659721","display_name":"Ahmad Tahmasivand","orcid":"https://orcid.org/0009-0005-5691-317X","observed_orcids":["https://orcid.org/0009-0005-5691-317X"]},"institutions":[{"id":"https://openalex.org/I162714631","display_name":"George Mason University","ror":"https://ror.org/02jqj7156","country_code":"US","type":"education","lineage":["https://openalex.org/I162714631"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ahmad Tahmasivand","raw_affiliation_strings":["Electrical and Computer Engineering Department, George Mason University, FairFax, VA, USA"],"raw_orcid":"https://orcid.org/0009-0005-5691-317X","affiliations":[{"raw_affiliation_string":"Electrical and Computer Engineering Department, George Mason University, FairFax, VA, USA","institution_ids":["https://openalex.org/I162714631"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5002123043","display_name":"Ihsen Alouani","orcid":"https://orcid.org/0000-0001-5102-8087","observed_orcids":["https://orcid.org/0000-0001-5102-8087"]},"institutions":[{"id":"https://openalex.org/I126231945","display_name":"Queen's University Belfast","ror":"https://ror.org/00hswnk62","country_code":"GB","type":"education","lineage":["https://openalex.org/I126231945"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Ihsen Alouani","raw_affiliation_strings":["School of Electronics, Electrical Engineering and Computer Science, Queen's University Belfast, Belfast, United Kingdom"],"raw_orcid":"https://orcid.org/0000-0001-5102-8087","affiliations":[{"raw_affiliation_string":"School of Electronics, Electrical Engineering and Computer Science, Queen's University Belfast, Belfast, United Kingdom","institution_ids":["https://openalex.org/I126231945"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5037831461","display_name":"Khaled N. Khasawneh","orcid":"https://orcid.org/0000-0002-2116-2223","observed_orcids":["https://orcid.org/0000-0002-2116-2223"]},"institutions":[{"id":"https://openalex.org/I162714631","display_name":"George Mason University","ror":"https://ror.org/02jqj7156","country_code":"US","type":"education","lineage":["https://openalex.org/I162714631"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Khaled Khasawneh","raw_affiliation_strings":["Electrical and Computer Engineering Department, George Mason University, FairFax, VA, USA"],"raw_orcid":"https://orcid.org/0000-0002-2116-2223","affiliations":[{"raw_affiliation_string":"Electrical and Computer Engineering Department, George Mason University, FairFax, VA, USA","institution_ids":["https://openalex.org/I162714631"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5084829415","display_name":"Mohammed E. Fouda","orcid":"https://orcid.org/0000-0001-7139-3428","observed_orcids":["https://orcid.org/0000-0001-7139-3428"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Mohammed Fouda","raw_affiliation_strings":["Compumacy for Artificial Intelligence solutions,, Cairo, Egypt"],"raw_orcid":"https://orcid.org/0000-0001-7139-3428","affiliations":[{"raw_affiliation_string":"Compumacy for Artificial Intelligence solutions,, Cairo, Egypt","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":3.9636,"has_fulltext":true,"cited_by_count":4,"citation_normalized_percentile":{"value":0.93944694,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":90,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"554","last_page":"561"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9990000128746033,"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":0.9990000128746033,"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/T11424","display_name":"Security and Verification in Computing","score":0.0003000000142492354,"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/T10951","display_name":"Cryptographic Implementations and Security","score":0.00019999999494757503,"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/jailbreaking","display_name":"jailbreaking","score":0.9179999828338623},{"id":"https://openalex.org/keywords/fault-injection-attacks","display_name":"fault injection attacks","score":0.8859999775886536},{"id":"https://openalex.org/keywords/quantized-language-models","display_name":"quantized language models","score":0.8130000233650208},{"id":"https://openalex.org/keywords/weight-only-quantization","display_name":"weight-only quantization","score":0.8629999756813049},{"id":"https://openalex.org/keywords/fp8","display_name":"FP8","score":0.6510000228881836},{"id":"https://openalex.org/keywords/int8","display_name":"INT8","score":0.8429999947547913},{"id":"https://openalex.org/keywords/attack-success-rate","display_name":"attack success rate","score":0.6069999933242798},{"id":"https://openalex.org/keywords/transferability","display_name":"transferability","score":0.5450000166893005},{"id":"https://openalex.org/keywords/model-integrity","display_name":"model integrity","score":0.3959999978542328},{"id":"https://openalex.org/keywords/safety-alignment","display_name":"safety alignment","score":0.5899999737739563}],"concepts":[{"id":"https://openalex.org/C2775928411","wikidata":"https://www.wikidata.org/wiki/Q2041312","display_name":"Fault injection","level":3,"score":0.6626626253128052},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6346487998962402},{"id":"https://openalex.org/C175551986","wikidata":"https://www.wikidata.org/wiki/Q47089","display_name":"Fault (geology)","level":2,"score":0.45899513363838196},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.36840903759002686},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.29755786061286926},{"id":"https://openalex.org/C127313418","wikidata":"https://www.wikidata.org/wiki/Q1069","display_name":"Geology","level":0,"score":0.134179949760437},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.0787833034992218},{"id":"https://openalex.org/C165205528","wikidata":"https://www.wikidata.org/wiki/Q83371","display_name":"Seismology","level":1,"score":0.06730493903160095}],"mesh":[],"locations_count":4,"locations":[{"id":"doi:10.1145/3716368.3735302","doi":"https://doi.org/10.1145/3716368.3735302","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3716368.3735302","pdf_url":null,"source":{"id":"https://openalex.org/S4363608736","display_name":"ACM Great Lakes Symposium on VLSI (GLSVLSI)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the Great Lakes Symposium on VLSI 2025","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2507.03236","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2507.03236","pdf_url":"https://arxiv.org/pdf/2507.03236","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"pmh:oai:pure.qub.ac.uk/portal:openaire/1b34d44f-6f4e-44f5-9dd1-87567e562153","doi":null,"is_oa":true,"landing_page_url":"https://pure.qub.ac.uk/en/publications/1b34d44f-6f4e-44f5-9dd1-87567e562153","pdf_url":"https://pureadmin.qub.ac.uk/ws/files/654033756/On_jailbreaking_quantized_Language_Models_through_fault_injection_attacks.pdf","source":{"id":"https://openalex.org/S4306402319","display_name":"Research Portal (Queen's University Belfast)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I126231945","host_organization_name":"Queen's University Belfast","host_organization_lineage":["https://openalex.org/I126231945"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Zahran, N, Tahmasivand, A, Alouani, I, Khasawneh, K N & Fouda, M 2025, On jailbreaking quantized Language Models through fault injection attacks. in L Peng, B Vaisband & F Chen (eds), GLSVLSI '25: Great Lakes Symposium on VLSI 2025: Proceedings. Association for Computing Machinery, pp. 554-561, GLSVLSI '25: Great Lakes Symposium on VLSI 2025, New Orleans, United States, 30/06/2025. https://doi.org/10.1145/3716368.3735302","raw_type":"info:eu-repo/semantics/conferenceObject"},{"id":"doi:10.48550/arxiv.2507.03236","doi":"https://doi.org/10.48550/arxiv.2507.03236","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2507.03236","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.1145/3716368.3735302","doi":"https://doi.org/10.1145/3716368.3735302","is_oa":true,"landing_page_url":"https://doi.org/10.1145/3716368.3735302","pdf_url":null,"source":{"id":"https://openalex.org/S4363608736","display_name":"ACM Great Lakes Symposium on VLSI (GLSVLSI)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by-nc-sa","license_id":"https://openalex.org/licenses/cc-by-nc-sa","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the Great Lakes Symposium on VLSI 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":10,"referenced_works":["https://openalex.org/W2157116240","https://openalex.org/W2979826702","https://openalex.org/W2981860227","https://openalex.org/W3030962502","https://openalex.org/W3106322382","https://openalex.org/W3199792160","https://openalex.org/W3202359271","https://openalex.org/W4396674231","https://openalex.org/W4399282312","https://openalex.org/W4404783422"],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2604133224","https://openalex.org/W2390279801","https://openalex.org/W3122756779","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109"],"abstract_inverted_index":{"The":[0],"safety":[1],"alignment":[2],"of":[3,45,118,162,219],"Language":[4],"Models":[5],"(LMs)":[6],"is":[7],"a":[8,61,70],"critical":[9],"concern,":[10],"yet":[11],"their":[12],"integrity":[13],"can":[14,225],"be":[15],"challenged":[16],"by":[17,27],"direct":[18,220],"parameter":[19,221],"manipulation":[20,222],"attacks,":[21,59],"such":[22,46],"as":[23],"those":[24],"potentially":[25],"induced":[26,183],"fault":[28],"injection.":[29],"As":[30],"LMs":[31,51],"are":[32],"increasingly":[33],"deployed":[34],"using":[35],"low-precision":[36],"quantization":[37,54,89,95,170,193,212],"for":[38,48],"efficiency,":[39],"this":[40],"paper":[41],"investigates":[42],"the":[43,133,217],"efficacy":[44],"attacks":[47,101],"jailbreaking":[49],"aligned":[50],"across":[52,84,169],"different":[53],"schemes.":[55],"We":[56],"propose":[57],"gradient-guided":[58],"including":[60],"tailored":[62],"progressive":[63],"bit-level":[64],"search":[65],"algorithm":[66],"introduced":[67],"herein":[68],"and":[69,82,87,122,129,152,175],"comparative":[71],"word-level":[72],"(single":[73],"weight":[74],"update)":[75],"attack.":[76],"Our":[77],"evaluation":[78],"on":[79,111],"Llama-3.2-3B,":[80],"Phi-4-mini,":[81],"Llama-3-8B":[83],"FP16":[85,112,185],"(baseline),":[86],"weight-only":[88],"(FP8,":[90],"INT8,":[91],"INT4)":[92,174],"reveals":[93],"that":[94,155,209],"significantly":[96,199],"influences":[97],"attack":[98,116],"success.":[99],"While":[100],"readily":[102],"achieve":[103],"high":[104,157],"success":[105],"(>80%":[106],"Attack":[107],"Success":[108],"Rate,":[109],"ASR)":[110],"models,":[113],"within":[114],"an":[115],"budget":[117,135],"25":[119],"perturbations,":[120],"FP8":[121,140],"INT8":[123,151],"models":[124,141,154,186],"exhibit":[125],"ASRs":[126],"below":[127,144],"20%":[128],"50%,":[130],"respectively.":[131],"Increasing":[132],"perturbation":[134,163],"up":[136],"to":[137,150,190],"150":[138],"bit-flips,":[139],"maintained":[142],"ASR":[143,195,202],"65%,":[145],"demonstrating":[146],"some":[147],"resilience":[148],"compared":[149],"INT4":[153,198],"have":[156],"ASR.":[158],"In":[159],"addition,":[160],"analysis":[161],"locations":[164],"revealed":[165],"differing":[166],"architectural":[167],"targets":[168],"schemes,":[171,213],"with":[172],"(FP16,":[173],"(INT8,":[176],"FP8)":[177],"showing":[178],"similar":[179],"characteristics.":[180],"Besides,":[181],"jailbreaks":[182],"in":[184],"were":[187],"highly":[188],"transferable":[189],"subsequent":[191],"FP8/INT8":[192],"(<5%":[194],"difference),":[196],"though":[197],"reduced":[200],"transferred":[201],"(avg.":[203],"35%":[204],"drop).":[205],"These":[206],"findings":[207],"highlight":[208],"while":[210],"common":[211],"particularly":[214],"FP8,":[215],"increase":[216],"difficulty":[218],"jailbreaks,":[223],"vulnerabilities":[224],"still":[226],"persist,":[227],"especially":[228],"through":[229],"post-attack":[230],"quantization.":[231]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":1}],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2025-10-10T00:00:00"}
