{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:IOMA4HTPOCXUEHCNDF742BFSNG","short_pith_number":"pith:IOMA4HTP","schema_version":"1.0","canonical_sha256":"43980e1e6f70af421c4d197fcd04b2698002d65720d8cd678b658ac3e36d08d1","source":{"kind":"arxiv","id":"2504.04377","version":2},"attestation_state":"computed","paper":{"title":"PolyGuard: A Multilingual Safety Moderation Tool for 17 Languages","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Akhila Yerukola, Devansh Jain, Himanshu Beniwal, Liwei Jiang, Maarten Sap, Priyanshu Kumar, Thomas Hartvigsen","submitted_at":"2025-04-06T06:09:21Z","abstract_excerpt":"Truly multilingual safety moderation efforts for Large Language Models (LLMs) have been hindered by a narrow focus on a small set of languages (e.g., English, Chinese) as well as a limited scope of safety definition, resulting in significant gaps in moderation capabilities. To bridge these gaps, we release POLYGUARD, a new state-of-the-art multilingual safety model for safeguarding LLM generations, and the corresponding training and evaluation datasets. POLYGUARD is trained on POLYGUARDMIX, the largest multilingual safety training corpus to date containing 1.91M samples across 17 languages (e."},"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":"2504.04377","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-04-06T06:09:21Z","cross_cats_sorted":[],"title_canon_sha256":"9aeb871d9a13149b24378dcb0c1b2de93e8e3d7df2ef33573862b8e2fa0000b8","abstract_canon_sha256":"85bfc32e482acd6c8adb26a1632599685591476d11178238ad9f7177e1ba3866"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:49:52.844407Z","signature_b64":"Lid7vxSQx30iL/1CdTYVhENgXxuNrF90VoSxsEKUC9MHizbQJ0mRzCmWTdmG3Bfi0KQi6Wk6Oc2ODRB+sxz1Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"43980e1e6f70af421c4d197fcd04b2698002d65720d8cd678b658ac3e36d08d1","last_reissued_at":"2026-07-05T11:49:52.843890Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:49:52.843890Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PolyGuard: A Multilingual Safety Moderation Tool for 17 Languages","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Akhila Yerukola, Devansh Jain, Himanshu Beniwal, Liwei Jiang, Maarten Sap, Priyanshu Kumar, Thomas Hartvigsen","submitted_at":"2025-04-06T06:09:21Z","abstract_excerpt":"Truly multilingual safety moderation efforts for Large Language Models (LLMs) have been hindered by a narrow focus on a small set of languages (e.g., English, Chinese) as well as a limited scope of safety definition, resulting in significant gaps in moderation capabilities. To bridge these gaps, we release POLYGUARD, a new state-of-the-art multilingual safety model for safeguarding LLM generations, and the corresponding training and evaluation datasets. POLYGUARD is trained on POLYGUARDMIX, the largest multilingual safety training corpus to date containing 1.91M samples across 17 languages (e."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.04377","kind":"arxiv","version":2},"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/2504.04377/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":"2504.04377","created_at":"2026-07-05T11:49:52.843954+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.04377v2","created_at":"2026-07-05T11:49:52.843954+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.04377","created_at":"2026-07-05T11:49:52.843954+00:00"},{"alias_kind":"pith_short_12","alias_value":"IOMA4HTPOCXU","created_at":"2026-07-05T11:49:52.843954+00:00"},{"alias_kind":"pith_short_16","alias_value":"IOMA4HTPOCXUEHCN","created_at":"2026-07-05T11:49:52.843954+00:00"},{"alias_kind":"pith_short_8","alias_value":"IOMA4HTP","created_at":"2026-07-05T11:49:52.843954+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06326","citing_title":"DT-Guard: Intent-Driven Reasoning-Active Training for Reasoning-Free LLM Safety Guardrail","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2606.25380","citing_title":"A Survey of Toxicity Detection and Mitigation Strategies for Multilingual Language Models","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2607.02079","citing_title":"HaloGuard 1.0: An Open Weights Constitutional Classifier for Multilingual AI Safety","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03647","citing_title":"Black-box, Adaptive, Efficient, Transferable, Harmful, Applicable... Attacks Are All You Need to Break LLMs","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.27997","citing_title":"Where Does Toxicity Live? Mechanistic Localization and Targeted Suppression in Language Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29659","citing_title":"Opir: Efficient Multi-Task Safety Classification for Toxicity, Jailbreaks, Hate Speech, and Harmful Content","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17329","citing_title":"LPG: Balancing Efficiency and Policy Reasoning in Latent Policy Guardrails","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05277","citing_title":"GLiNER Guard: Unified Encoder Family for Production LLM Safety and Privacy","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07982","citing_title":"GLiGuard: Schema-Conditioned Classification for LLM Safeguard","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18519","citing_title":"LLM Safety From Within: Detecting Harmful Content with Internal Representations","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16542","citing_title":"TWGuard: A Case Study of LLM Safety Guardrails for Localized Linguistic Contexts","ref_index":8,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG","json":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG.json","graph_json":"https://pith.science/api/pith-number/IOMA4HTPOCXUEHCNDF742BFSNG/graph.json","events_json":"https://pith.science/api/pith-number/IOMA4HTPOCXUEHCNDF742BFSNG/events.json","paper":"https://pith.science/paper/IOMA4HTP"},"agent_actions":{"view_html":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG","download_json":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG.json","view_paper":"https://pith.science/paper/IOMA4HTP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.04377&json=true","fetch_graph":"https://pith.science/api/pith-number/IOMA4HTPOCXUEHCNDF742BFSNG/graph.json","fetch_events":"https://pith.science/api/pith-number/IOMA4HTPOCXUEHCNDF742BFSNG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG/action/storage_attestation","attest_author":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG/action/author_attestation","sign_citation":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG/action/citation_signature","submit_replication":"https://pith.science/pith/IOMA4HTPOCXUEHCNDF742BFSNG/action/replication_record"}},"created_at":"2026-07-05T11:49:52.843954+00:00","updated_at":"2026-07-05T11:49:52.843954+00:00"}