{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:2C5FQKYPGW24CFMA4OGXYFHUNI","short_pith_number":"pith:2C5FQKYP","schema_version":"1.0","canonical_sha256":"d0ba582b0f35b5c11580e38d7c14f46a19141ba58e5252d5433334f7f24ce5cc","source":{"kind":"arxiv","id":"2310.02446","version":2},"attestation_state":"computed","paper":{"title":"Low-Resource Languages Jailbreak GPT-4","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Translating harmful English prompts into low-resource languages lets GPT-4 provide actionable advice for bad goals 79 percent of the time.","cross_cats":["cs.AI","cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Cristina Menghini, Stephen H. Bach, Zheng-Xin Yong","submitted_at":"2023-10-03T21:30:56Z","abstract_excerpt":"AI safety training and red-teaming of large language models (LLMs) are measures to mitigate the generation of unsafe content. Our work exposes the inherent cross-lingual vulnerability of these safety mechanisms, resulting from the linguistic inequality of safety training data, by successfully circumventing GPT-4's safeguard through translating unsafe English inputs into low-resource languages. On the AdvBenchmark, GPT-4 engages with the unsafe translated inputs and provides actionable items that can get the users towards their harmful goals 79% of the time, which is on par with or even surpass"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2310.02446","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-10-03T21:30:56Z","cross_cats_sorted":["cs.AI","cs.CR","cs.LG"],"title_canon_sha256":"33bbf057176543a71f37ad4685c2ce1ea6fbd47b59450069528b23f272988709","abstract_canon_sha256":"30e80a9acc7c61f31bb9748fe5e9274c34733005765d8178bff8e059c3b3b223"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:38:14.486936Z","signature_b64":"2h8Mb/uDfoxfPSBTjg5RX/HTGzB5mofwKXE4U2pHdxlloHjEhyNK3GwHFc1FpAHQRQCBHIu6DYBoyvATj9mTDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d0ba582b0f35b5c11580e38d7c14f46a19141ba58e5252d5433334f7f24ce5cc","last_reissued_at":"2026-05-17T23:38:14.486235Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:38:14.486235Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Low-Resource Languages Jailbreak GPT-4","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Translating harmful English prompts into low-resource languages lets GPT-4 provide actionable advice for bad goals 79 percent of the time.","cross_cats":["cs.AI","cs.CR","cs.LG"],"primary_cat":"cs.CL","authors_text":"Cristina Menghini, Stephen H. Bach, Zheng-Xin Yong","submitted_at":"2023-10-03T21:30:56Z","abstract_excerpt":"AI safety training and red-teaming of large language models (LLMs) are measures to mitigate the generation of unsafe content. Our work exposes the inherent cross-lingual vulnerability of these safety mechanisms, resulting from the linguistic inequality of safety training data, by successfully circumventing GPT-4's safeguard through translating unsafe English inputs into low-resource languages. On the AdvBenchmark, GPT-4 engages with the unsafe translated inputs and provides actionable items that can get the users towards their harmful goals 79% of the time, which is on par with or even surpass"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On the AdvBenchmark, GPT-4 engages with the unsafe translated inputs and provides actionable items that can get the users towards their harmful goals 79% of the time, which is on par with or even surpassing state-of-the-art jailbreaking attacks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That automatic translations preserve the original unsafe intent without introducing detectable artifacts or triggering the model's cross-lingual safety mechanisms, and that results on the tested benchmark generalize to other prompts and models.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Translating unsafe inputs to low-resource languages jailbreaks GPT-4 at rates on par with or exceeding state-of-the-art attacks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Translating harmful English prompts into low-resource languages lets GPT-4 provide actionable advice for bad goals 79 percent of the time.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5acc8099a82390bf1e378b69ad7a2679c90d48e86db7de89c166a8b17df892d1"},"source":{"id":"2310.02446","kind":"arxiv","version":2},"verdict":{"id":"fe7316e4-a497-4337-b3d2-1f0fc27eff95","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-17T09:12:43.135443Z","strongest_claim":"On the AdvBenchmark, GPT-4 engages with the unsafe translated inputs and provides actionable items that can get the users towards their harmful goals 79% of the time, which is on par with or even surpassing state-of-the-art jailbreaking attacks.","one_line_summary":"Translating unsafe inputs to low-resource languages jailbreaks GPT-4 at rates on par with or exceeding state-of-the-art attacks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That automatic translations preserve the original unsafe intent without introducing detectable artifacts or triggering the model's cross-lingual safety mechanisms, and that results on the tested benchmark generalize to other prompts and models.","pith_extraction_headline":"Translating harmful English prompts into low-resource languages lets GPT-4 provide actionable advice for bad goals 79 percent of the time."},"references":{"count":55,"sample":[{"doi":"","year":2020,"title":"Jigsaw multilingual toxic comment classification, 2020","work_id":"acd2de50-2c59-482c-82e8-153db57bc84d","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback","work_id":"a1f2574b-a899-4713-be60-c87ba332656c","ref_index":2,"cited_arxiv_id":"2204.05862","is_internal_anchor":true},{"doi":"","year":2022,"title":"Constitutional AI: Harmlessness from AI Feedback","work_id":"faaaa4e0-2676-4fac-a0b4-99aef10d2095","ref_index":3,"cited_arxiv_id":"2212.08073","is_internal_anchor":true},{"doi":"","year":2023,"title":"A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity","work_id":"41dff74c-00b2-4c77-a674-9f86030c06c8","ref_index":4,"cited_arxiv_id":"2302.04023","is_internal_anchor":true},{"doi":"","year":2022,"title":"X., Macherey, K., Krikun, M., Wang, P., Gutkin, A., Shah, A., Huang, Y., Chen, Z., Wu, Y., and Hughes, M","work_id":"c55757ac-fd65-4628-8d13-7ce5b6c20ff0","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":55,"snapshot_sha256":"7c923c5d51f1b6baab3a00891cd1dffa7a58bf223887463a1b27ff32553e91a3","internal_anchors":13},"formal_canon":{"evidence_count":2,"snapshot_sha256":"685cae991da7393e3601bb01782359bffbed4eef16ae8255aba754e370d4704d"},"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":"2310.02446","created_at":"2026-05-17T23:38:14.486366+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.02446v2","created_at":"2026-05-17T23:38:14.486366+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.02446","created_at":"2026-05-17T23:38:14.486366+00:00"},{"alias_kind":"pith_short_12","alias_value":"2C5FQKYPGW24","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_16","alias_value":"2C5FQKYPGW24CFMA","created_at":"2026-05-18T12:33:33.725879+00:00"},{"alias_kind":"pith_short_8","alias_value":"2C5FQKYP","created_at":"2026-05-18T12:33:33.725879+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":20,"internal_anchor_count":20,"sample":[{"citing_arxiv_id":"2511.02356","citing_title":"ASTRA: An Automated Framework for Strategy Discovery, Retrieval, and Evolution for Jailbreaking LLMs","ref_index":54,"is_internal_anchor":true},{"citing_arxiv_id":"2402.10260","citing_title":"A StrongREJECT for Empty Jailbreaks","ref_index":39,"is_internal_anchor":true},{"citing_arxiv_id":"2404.01318","citing_title":"JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models","ref_index":56,"is_internal_anchor":true},{"citing_arxiv_id":"2407.04295","citing_title":"Jailbreak Attacks and Defenses Against Large Language Models: A Survey","ref_index":106,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12876","citing_title":"Certified Robustness under Heterogeneous Perturbations via Hybrid Randomized Smoothing","ref_index":17,"is_internal_anchor":true},{"citing_arxiv_id":"2310.03684","citing_title":"SmoothLLM: Defending Large Language Models Against Jailbreaking Attacks","ref_index":28,"is_internal_anchor":true},{"citing_arxiv_id":"2410.09024","citing_title":"AgentHarm: A Benchmark for Measuring Harmfulness of LLM Agents","ref_index":32,"is_internal_anchor":true},{"citing_arxiv_id":"2604.01473","citing_title":"SelfGrader: Stable Jailbreak Detection for Large Language Models using Token-Level Logits","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05630","citing_title":"One Turn Too Late: Response-Aware Defense Against Hidden Malicious Intent in Multi-Turn Dialogue","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10998","citing_title":"Few-Shot Truly Benign DPO Attack for Jailbreaking LLMs","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2605.04019","citing_title":"Redefining AI Red Teaming in the Agentic Era: From Weeks to Hours","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05630","citing_title":"One Turn Too Late: Response-Aware Defense Against Hidden Malicious Intent in Multi-Turn Dialogue","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2604.22427","citing_title":"Automation-Exploit: A Multi-Agent LLM Framework for Adaptive Offensive Security with Digital Twin-Based Risk-Mitigated Exploitation","ref_index":47,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01034","citing_title":"A Theoretical Game of Attacks via Compositional Skills","ref_index":16,"is_internal_anchor":true},{"citing_arxiv_id":"2605.00236","citing_title":"Attention Is Where You Attack","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2604.20720","citing_title":"COMPASS: COntinual Multilingual PEFT with Adaptive Semantic Sampling","ref_index":82,"is_internal_anchor":true},{"citing_arxiv_id":"2604.12710","citing_title":"LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2604.12232","citing_title":"TEMPLATEFUZZ: Fine-Grained Chat Template Fuzzing for Jailbreaking and Red Teaming LLMs","ref_index":59,"is_internal_anchor":true},{"citing_arxiv_id":"2605.02971","citing_title":"Multilingual Safety Alignment via Self-Distillation","ref_index":18,"is_internal_anchor":true},{"citing_arxiv_id":"2605.02971","citing_title":"Multilingual Safety Alignment via Self-Distillation","ref_index":18,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI","json":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI.json","graph_json":"https://pith.science/api/pith-number/2C5FQKYPGW24CFMA4OGXYFHUNI/graph.json","events_json":"https://pith.science/api/pith-number/2C5FQKYPGW24CFMA4OGXYFHUNI/events.json","paper":"https://pith.science/paper/2C5FQKYP"},"agent_actions":{"view_html":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI","download_json":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI.json","view_paper":"https://pith.science/paper/2C5FQKYP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.02446&json=true","fetch_graph":"https://pith.science/api/pith-number/2C5FQKYPGW24CFMA4OGXYFHUNI/graph.json","fetch_events":"https://pith.science/api/pith-number/2C5FQKYPGW24CFMA4OGXYFHUNI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI/action/storage_attestation","attest_author":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI/action/author_attestation","sign_citation":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI/action/citation_signature","submit_replication":"https://pith.science/pith/2C5FQKYPGW24CFMA4OGXYFHUNI/action/replication_record"}},"created_at":"2026-05-17T23:38:14.486366+00:00","updated_at":"2026-05-17T23:38:14.486366+00:00"}