{"id":"https://openalex.org/W7225399035","doi":"https://doi.org/10.21203/rs.3.rs-11242446/v1","title":"Red-Teaming African-Language LLMs: Safety Alignment Failures Across Low-Resource Languages","display_name":"Red-Teaming African-Language LLMs: Safety Alignment Failures Across Low-Resource Languages","publication_year":2026,"publication_date":"2026-11-06","ids":{"openalex":"https://openalex.org/W7225399035","doi":"https://doi.org/10.21203/rs.3.rs-11242446/v1"},"language":"en","primary_location":{"id":"doi:10.21203/rs.3.rs-11242446/v1","doi":"https://doi.org/10.21203/rs.3.rs-11242446/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-11242446/v1","pdf_url":"https://www.researchsquare.com/article/rs-11242446/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","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":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://www.researchsquare.com/article/rs-11242446/latest.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5161032420","display_name":"Godwill Nkwain Nyong","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210130200","display_name":"Carnegie Mellon University Africa","ror":"https://ror.org/02f33m021","country_code":"RW","type":"education","lineage":["https://openalex.org/I4210130200","https://openalex.org/I74973139"]},{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["RW","US"],"is_corresponding":false,"raw_author_name":"Godwill Nkwain Nyong","raw_affiliation_strings":["Carnegie Mellon University Africa"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University Africa","institution_ids":["https://openalex.org/I4210130200","https://openalex.org/I74973139"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060594789","display_name":"George Onyango Okeyo","orcid":"https://orcid.org/0000-0003-2276-2260","observed_orcids":["https://orcid.org/0000-0003-2276-2260"]},"institutions":[{"id":"https://openalex.org/I4210130200","display_name":"Carnegie Mellon University Africa","ror":"https://ror.org/02f33m021","country_code":"RW","type":"education","lineage":["https://openalex.org/I4210130200","https://openalex.org/I74973139"]},{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["RW","US"],"is_corresponding":false,"raw_author_name":"George Okeyo","raw_affiliation_strings":["Carnegie Mellon University Africa"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University Africa","institution_ids":["https://openalex.org/I4210130200","https://openalex.org/I74973139"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5034886864","display_name":"Jema David Ndibwile","orcid":"https://orcid.org/0000-0002-7962-2237","observed_orcids":["https://orcid.org/0000-0002-7962-2237"]},"institutions":[{"id":"https://openalex.org/I4210130200","display_name":"Carnegie Mellon University Africa","ror":"https://ror.org/02f33m021","country_code":"RW","type":"education","lineage":["https://openalex.org/I4210130200","https://openalex.org/I74973139"]},{"id":"https://openalex.org/I74973139","display_name":"Carnegie Mellon University","ror":"https://ror.org/05x2bcf33","country_code":"US","type":"education","lineage":["https://openalex.org/I74973139"]}],"countries":["RW","US"],"is_corresponding":false,"raw_author_name":"Jema David Ndibwile","raw_affiliation_strings":["Carnegie Mellon University Africa"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Carnegie Mellon University Africa","institution_ids":["https://openalex.org/I4210130200","https://openalex.org/I74973139"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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":0.9951000213623047,"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":0.9951000213623047,"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/T12262","display_name":"Hate Speech and Cyberbullying Detection","score":0.0017999999690800905,"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":0.0012000000569969416,"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/african-languages","display_name":"African languages","score":0.6830000281333923},{"id":"https://openalex.org/keywords/large-language-model-safety","display_name":"large language model safety","score":0.753000020980835},{"id":"https://openalex.org/keywords/low-resource-languages","display_name":"low-resource languages","score":0.9139999747276306},{"id":"https://openalex.org/keywords/propaganda","display_name":"propaganda","score":0.5699999928474426},{"id":"https://openalex.org/keywords/hausa","display_name":"Hausa language","score":0.8830000162124634},{"id":"https://openalex.org/keywords/swahili-language","display_name":"Swahili language","score":0.9980000257492065},{"id":"https://openalex.org/keywords/yoruba-language","display_name":"Yoruba","score":0.9980000257492065},{"id":"https://openalex.org/keywords/igbo","display_name":"Igbo","score":1.0},{"id":"https://openalex.org/keywords/m-pesa","display_name":"M-Pesa","score":0.6449999809265137},{"id":"https://openalex.org/keywords/independent-national-electoral-commission","display_name":"Independent National Electoral Commission","score":0.6880000233650208},{"id":"https://openalex.org/keywords/safaricom","display_name":"Safaricom","score":0.9350000023841858},{"id":"https://openalex.org/keywords/defense-mechanisms","display_name":"defense mechanisms","score":0.460999995470047}],"concepts":[{"id":"https://openalex.org/C2777363581","wikidata":"https://www.wikidata.org/wiki/Q15098235","display_name":"Harm","level":2,"score":0.6951000094413757},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.6351000070571899},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6243000030517578},{"id":"https://openalex.org/C95713431","wikidata":"https://www.wikidata.org/wiki/Q631425","display_name":"Vulnerability (computing)","level":2,"score":0.45570001006126404},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.43939998745918274},{"id":"https://openalex.org/C58642233","wikidata":"https://www.wikidata.org/wiki/Q8269924","display_name":"Taxonomy (biology)","level":2,"score":0.41370001435279846},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.39489999413490295},{"id":"https://openalex.org/C2776034101","wikidata":"https://www.wikidata.org/wiki/Q1509347","display_name":"Commission","level":2,"score":0.37279999256134033},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.35510000586509705},{"id":"https://openalex.org/C2778571376","wikidata":"https://www.wikidata.org/wiki/Q1355821","display_name":"Frontier","level":2,"score":0.35019999742507935},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.298799991607666},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2987000048160553},{"id":"https://openalex.org/C2910001868","wikidata":"https://www.wikidata.org/wiki/Q458","display_name":"European union","level":2,"score":0.29280000925064087},{"id":"https://openalex.org/C3020782553","wikidata":"https://www.wikidata.org/wiki/Q8880","display_name":"European commission","level":3,"score":0.29120001196861267},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.28839999437332153},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.28760001063346863},{"id":"https://openalex.org/C44725695","wikidata":"https://www.wikidata.org/wiki/Q288156","display_name":"Normative","level":2,"score":0.27810001373291016},{"id":"https://openalex.org/C31170391","wikidata":"https://www.wikidata.org/wiki/Q188619","display_name":"Hierarchy","level":2,"score":0.2773999869823456},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.27649998664855957},{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.2718000113964081},{"id":"https://openalex.org/C153924320","wikidata":"https://www.wikidata.org/wiki/Q56475","display_name":"Hausa","level":2,"score":0.2624000012874603},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2619999945163727},{"id":"https://openalex.org/C2522767166","wikidata":"https://www.wikidata.org/wiki/Q2374463","display_name":"Data science","level":1,"score":0.26100000739097595},{"id":"https://openalex.org/C42475967","wikidata":"https://www.wikidata.org/wiki/Q194292","display_name":"Operations research","level":1,"score":0.25769999623298645},{"id":"https://openalex.org/C60643870","wikidata":"https://www.wikidata.org/wiki/Q1949683","display_name":"Deterrence theory","level":2,"score":0.2558000087738037}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.21203/rs.3.rs-11242446/v1","doi":"https://doi.org/10.21203/rs.3.rs-11242446/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-11242446/v1","pdf_url":"https://www.researchsquare.com/article/rs-11242446/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","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":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"}],"best_oa_location":{"id":"doi:10.21203/rs.3.rs-11242446/v1","doi":"https://doi.org/10.21203/rs.3.rs-11242446/v1","is_oa":true,"landing_page_url":"https://doi.org/10.21203/rs.3.rs-11242446/v1","pdf_url":"https://www.researchsquare.com/article/rs-11242446/latest.pdf","source":{"id":"https://openalex.org/S4306525896","display_name":"Research Square","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":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Research Square","raw_type":"posted-content"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7225399035.pdf","grobid_xml":"https://content.openalex.org/works/W7225399035.grobid-xml"},"referenced_works_count":14,"referenced_works":["https://openalex.org/W3207937903","https://openalex.org/W4223908421","https://openalex.org/W4283170666","https://openalex.org/W4296413526","https://openalex.org/W4366735819","https://openalex.org/W4385574228","https://openalex.org/W4385894687","https://openalex.org/W4387324089","https://openalex.org/W4387561119","https://openalex.org/W4394972055","https://openalex.org/W4404782984","https://openalex.org/W4411119839","https://openalex.org/W7133208539","https://openalex.org/W7133211372"],"related_works":["https://openalex.org/W7224922142","https://openalex.org/W7224097909","https://openalex.org/W7163218506","https://openalex.org/W7140118283","https://openalex.org/W7214914238","https://openalex.org/W7223053443","https://openalex.org/W7204198567","https://openalex.org/W4411119067","https://openalex.org/W4416033559","https://openalex.org/W7138888256"],"abstract_inverted_index":{"Abstract":[0],"Large":[1],"Language":[2],"Models":[3],"are":[4,167],"now":[5],"being":[6],"developed":[7],"specifically":[8],"for":[9,40],"African":[10,114],"languages,":[11],"with":[12,146],"models":[13,42,59,82,198],"like":[14,34],"N-ATLaS":[15],"and":[16,36,99,123,134,163,196],"Pawa-Gemma-Swahili":[17],"achieving":[18],"meaningful":[19],"capability":[20,28],"gains":[21],"on":[22,52],"standard":[23],"NLP":[24],"benchmarks.":[25],"Yet":[26],"while":[27,170],"evaluation":[29,39,74,227],"has":[30],"advanced":[31],"through":[32],"benchmarks":[33],"IrokoBench":[35],"AfroBench,":[37],"safety":[38,49],"these":[41],"remains":[43],"less":[44],"or":[45],"nonexistent.":[46],"Existing":[47],"multilingual":[48],"research":[50],"focuses":[51],"using":[53,106],"low-resource":[54],"languages":[55,85,149],"to":[56,222],"attack":[57,110],"frontier":[58],"such":[60],"as":[61],"GPT-4,":[62],"leaving":[63],"African-language":[64,76,232],"LLMs":[65],"themselves":[66],"unexamined.":[67],"We":[68],"present":[69],"the":[70,173,194,203],"first":[71],"systematic":[72],"red-team":[73],"of":[75,141],"LLMs,":[77],"testing":[78],"two":[79],"publicly":[80],"available":[81],"across":[83],"four":[84,95,148],"(Hausa,":[86],"Swahili,":[87],"Yoruba,":[88],"Igbo),":[89],"eight":[90],"harm":[91],"categories":[92],"organized":[93],"into":[94],"HEVP":[96,156],"threat":[97],"domains,":[98],"ten":[100],"adversarial":[101],"techniques.":[102],"Across":[103],"220":[104],"trials":[105],"24":[107],"culturally":[108],"grounded":[109],"configurations":[111],"referencing":[112],"real":[113],"systems":[115],"(M-Pesa":[116],"mobile":[117],"money,":[118],"Nigeria's":[119,128],"National":[120,130],"Identification":[121],"Number":[122,126],"Bank":[124],"Verification":[125],"(NIN/BVN),":[127],"Independent":[129],"Electoral":[131],"Commission":[132],"(INEC),":[133],"Safaricom":[135],"telecommunications),":[136],"we":[137],"find":[138],"that":[139],"44.1%":[140],"attacks":[142],"elicit":[143],"harmful":[144],"content,":[145],"all":[147,184],"vulnerable":[150],"at":[151],"comparable":[152],"rates":[153],"(40.0--48.0%).":[154],"The":[155],"taxonomy":[157],"reveals":[158],"structured":[159],"vulnerability:":[160],"Propaganda":[161],"(70.0%)":[162],"Harm":[164],"Generation":[165],"(58.0%)":[166],"most":[168],"exploitable,":[169],"hate/harassment":[171],"shows":[172],"strongest":[174],"defense":[175],"(12.0%).":[176],"Role":[177],"Exploitation":[178],"techniques":[179],"achieve":[180],"82.5%":[181],"success,":[182],"outperforming":[183],"other":[185],"technique":[186],"classes.":[187],"Model":[188],"size":[189],"does":[190],"not":[191],"predict":[192],"safety;":[193],"8B-parameter":[195],"2.6B-parameter":[197],"show":[199],"equivalent":[200],"vulnerability":[201],"despite":[202],"smaller":[204],"model":[205,233],"incorporating":[206],"Direct":[207],"Preference":[208],"Optimization":[209],"(DPO)":[210],"alignment.":[211],"Defense":[212],"mechanisms":[213],"vary":[214],"qualitatively":[215],"by":[216],"language,":[217],"from":[218],"disclaimers":[219],"in":[220,224,231],"Hausa":[221],"truncation":[223],"Swahili.":[225],"Safety":[226],"should":[228],"become":[229],"routine":[230],"development.":[234]},"counts_by_year":[],"updated_date":"2026-11-09T13:22:03.289127","created_date":"2026-11-07T00:00:00"}