{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GTEQAZMKASOZELVW5O5QBYETHS","short_pith_number":"pith:GTEQAZMK","schema_version":"1.0","canonical_sha256":"34c900658a049d922eb6ebbb00e0933cbe9df7b3adcf37c0aeea942483090a40","source":{"kind":"arxiv","id":"2404.19597","version":3},"attestation_state":"computed","paper":{"title":"TuBA: Cross-Lingual Transferability of Backdoor Attacks in LLMs with Instruction Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.CL","authors_text":"Benjamin I. P. Rubinstein, Jun Wang, Pasquale Minervini, Pontus Stenetorp, Qiongkai Xu, Trevor Cohn, Xuanli He","submitted_at":"2024-04-30T14:43:57Z","abstract_excerpt":"The implications of backdoor attacks on English-centric large language models (LLMs) have been widely examined - such attacks can be achieved by embedding malicious behaviors during training and activated under specific conditions that trigger malicious outputs. Despite the increasing support for multilingual capabilities in open-source and proprietary LLMs, the impact of backdoor attacks on these systems remains largely under-explored. Our research focuses on cross-lingual backdoor attacks against multilingual LLMs, particularly investigating how poisoning the instruction-tuning data for one "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2404.19597","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-04-30T14:43:57Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"07c77242bf439bc3016b56b2acc5ef6bc625ce995f4ac0c182ba8ea2b3804392","abstract_canon_sha256":"d96d8dcb0883122fda0ef2ba02261c332a74c7795c9d4fda947f361f791257a0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:32:26.448498Z","signature_b64":"wqgsYJdr891p4VthzL4D1LxFeMZp28cVInLBNDLFta1YnMyVfhrSampoWTMP+LYYvDhku7oM+sPEbtl6MTHYAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"34c900658a049d922eb6ebbb00e0933cbe9df7b3adcf37c0aeea942483090a40","last_reissued_at":"2026-07-05T10:32:26.447460Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:32:26.447460Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TuBA: Cross-Lingual Transferability of Backdoor Attacks in LLMs with Instruction Tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.CL","authors_text":"Benjamin I. P. Rubinstein, Jun Wang, Pasquale Minervini, Pontus Stenetorp, Qiongkai Xu, Trevor Cohn, Xuanli He","submitted_at":"2024-04-30T14:43:57Z","abstract_excerpt":"The implications of backdoor attacks on English-centric large language models (LLMs) have been widely examined - such attacks can be achieved by embedding malicious behaviors during training and activated under specific conditions that trigger malicious outputs. Despite the increasing support for multilingual capabilities in open-source and proprietary LLMs, the impact of backdoor attacks on these systems remains largely under-explored. Our research focuses on cross-lingual backdoor attacks against multilingual LLMs, particularly investigating how poisoning the instruction-tuning data for one "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.19597","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2404.19597/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2404.19597","created_at":"2026-07-05T10:32:26.447592+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.19597v3","created_at":"2026-07-05T10:32:26.447592+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.19597","created_at":"2026-07-05T10:32:26.447592+00:00"},{"alias_kind":"pith_short_12","alias_value":"GTEQAZMKASOZ","created_at":"2026-07-05T10:32:26.447592+00:00"},{"alias_kind":"pith_short_16","alias_value":"GTEQAZMKASOZELVW","created_at":"2026-07-05T10:32:26.447592+00:00"},{"alias_kind":"pith_short_8","alias_value":"GTEQAZMK","created_at":"2026-07-05T10:32:26.447592+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2508.10404","citing_title":"Layer-Wise Perturbations via Sparse Autoencoders for Adversarial Text Generation","ref_index":173,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS","json":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS.json","graph_json":"https://pith.science/api/pith-number/GTEQAZMKASOZELVW5O5QBYETHS/graph.json","events_json":"https://pith.science/api/pith-number/GTEQAZMKASOZELVW5O5QBYETHS/events.json","paper":"https://pith.science/paper/GTEQAZMK"},"agent_actions":{"view_html":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS","download_json":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS.json","view_paper":"https://pith.science/paper/GTEQAZMK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.19597&json=true","fetch_graph":"https://pith.science/api/pith-number/GTEQAZMKASOZELVW5O5QBYETHS/graph.json","fetch_events":"https://pith.science/api/pith-number/GTEQAZMKASOZELVW5O5QBYETHS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS/action/storage_attestation","attest_author":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS/action/author_attestation","sign_citation":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS/action/citation_signature","submit_replication":"https://pith.science/pith/GTEQAZMKASOZELVW5O5QBYETHS/action/replication_record"}},"created_at":"2026-07-05T10:32:26.447592+00:00","updated_at":"2026-07-05T10:32:26.447592+00:00"}