{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:PJ5C5SWJ3U4KWC3EQZ5BXUG4NO","short_pith_number":"pith:PJ5C5SWJ","schema_version":"1.0","canonical_sha256":"7a7a2ecac9dd38ab0b64867a1bd0dc6bae7df4afcf3dbcc6914fe97f034a6d46","source":{"kind":"arxiv","id":"2602.03328","version":2},"attestation_state":"computed","paper":{"title":"GuardReasoner-Omni: A Reasoning-based Multi-modal Guardrail for Text, Image, Video, and Audio","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Cancan Chen, Huiying Xu, Jiaheng Zhang, Tianyi Wu, Wenjie Qu, Xinzhong Zhu, Yanpei Guo, Yibo Li, Yue Liu, Yufei He, Yulin Chen, Zhenhao Zhu","submitted_at":"2026-02-03T09:56:20Z","abstract_excerpt":"We present GuardReasoner-Omni, a reasoning-based guardrail model designed to moderate text, image, video, and audio data. First, we construct a comprehensive training corpus comprising 181k samples spanning these four modalities. Our training pipeline follows a two-stage paradigm to incentivize the model to deliberate before making decisions: (1) conducting SFT to cold-start the model with explicit reasoning capabilities and structural adherence; and (2) performing RL with a concise correctness reward to preserve accurate reasoning while suppressing redundant generation. We release a suite of "},"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":"2602.03328","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2026-02-03T09:56:20Z","cross_cats_sorted":[],"title_canon_sha256":"51e7a856c729dd89c11d68aaa9c42a7cc5d5940b80bb901641140557259a334a","abstract_canon_sha256":"8b87383fa3ba285595d33ba973591ed6ab53ff2dd98c85e80c4778e254ca1136"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:05:11.706829Z","signature_b64":"YLnrYxqt0RTAL7pJXvT9uEQS3CiXUf7/CH+AyNNYcyqezuhFpby4Fw4vL6VKZU1vJLetghVgVQwTR3D4AuNmBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7a7a2ecac9dd38ab0b64867a1bd0dc6bae7df4afcf3dbcc6914fe97f034a6d46","last_reissued_at":"2026-05-28T01:05:11.706356Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:05:11.706356Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GuardReasoner-Omni: A Reasoning-based Multi-modal Guardrail for Text, Image, Video, and Audio","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Cancan Chen, Huiying Xu, Jiaheng Zhang, Tianyi Wu, Wenjie Qu, Xinzhong Zhu, Yanpei Guo, Yibo Li, Yue Liu, Yufei He, Yulin Chen, Zhenhao Zhu","submitted_at":"2026-02-03T09:56:20Z","abstract_excerpt":"We present GuardReasoner-Omni, a reasoning-based guardrail model designed to moderate text, image, video, and audio data. First, we construct a comprehensive training corpus comprising 181k samples spanning these four modalities. Our training pipeline follows a two-stage paradigm to incentivize the model to deliberate before making decisions: (1) conducting SFT to cold-start the model with explicit reasoning capabilities and structural adherence; and (2) performing RL with a concise correctness reward to preserve accurate reasoning while suppressing redundant generation. We release a suite of "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.03328","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/2602.03328/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":"2602.03328","created_at":"2026-05-28T01:05:11.706420+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.03328v2","created_at":"2026-05-28T01:05:11.706420+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.03328","created_at":"2026-05-28T01:05:11.706420+00:00"},{"alias_kind":"pith_short_12","alias_value":"PJ5C5SWJ3U4K","created_at":"2026-05-28T01:05:11.706420+00:00"},{"alias_kind":"pith_short_16","alias_value":"PJ5C5SWJ3U4KWC3E","created_at":"2026-05-28T01:05:11.706420+00:00"},{"alias_kind":"pith_short_8","alias_value":"PJ5C5SWJ","created_at":"2026-05-28T01:05:11.706420+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.17610","citing_title":"SafeLens: Deliberate and Efficient Video Guardrails with Fast-and-Slow Screening","ref_index":24,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO","json":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO.json","graph_json":"https://pith.science/api/pith-number/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/graph.json","events_json":"https://pith.science/api/pith-number/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/events.json","paper":"https://pith.science/paper/PJ5C5SWJ"},"agent_actions":{"view_html":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO","download_json":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO.json","view_paper":"https://pith.science/paper/PJ5C5SWJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.03328&json=true","fetch_graph":"https://pith.science/api/pith-number/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/graph.json","fetch_events":"https://pith.science/api/pith-number/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/action/storage_attestation","attest_author":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/action/author_attestation","sign_citation":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/action/citation_signature","submit_replication":"https://pith.science/pith/PJ5C5SWJ3U4KWC3EQZ5BXUG4NO/action/replication_record"}},"created_at":"2026-05-28T01:05:11.706420+00:00","updated_at":"2026-05-28T01:05:11.706420+00:00"}