{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:HFLUPVATQKVHOBNS4ESZWLBRVJ","short_pith_number":"pith:HFLUPVAT","schema_version":"1.0","canonical_sha256":"395747d41382aa7705b2e1259b2c31aa7270eb7e0b823987aa391d8a129532b4","source":{"kind":"arxiv","id":"2504.01081","version":2},"attestation_state":"computed","paper":{"title":"ShieldGemma 2: Robust and Tractable Image Content Moderation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","eess.IV"],"primary_cat":"cs.CV","authors_text":"Aparna Joshi, Cai Xu, Dana Kurniawan, Dirichi Ike-Njoku, Hamid Palangi, Jindong Gu, Jingjing Zhou, Joon Baek, Karthik Narasimhan, Mani Malek, Rick Pereira, Ryan Mullins, Shravan Dheep, Tamoghna Saha, Wenjun Zeng, Yiwen Song, Yuchi Liu","submitted_at":"2025-04-01T18:00:20Z","abstract_excerpt":"We introduce ShieldGemma 2, a 4B parameter image content moderation model built on Gemma 3. This model provides robust safety risk predictions across the following key harm categories: Sexually Explicit, Violence \\& Gore, and Dangerous Content for synthetic images (e.g. output of any image generation model) and natural images (e.g. any image input to a Vision-Language Model). We evaluated on both internal and external benchmarks to demonstrate state-of-the-art performance compared to LlavaGuard \\citep{helff2024llavaguard}, GPT-4o mini \\citep{hurst2024gpt}, and the base Gemma 3 model \\citep{gem"},"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":"2504.01081","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-04-01T18:00:20Z","cross_cats_sorted":["cs.CL","eess.IV"],"title_canon_sha256":"43809eadebac94dd9935f3fa83480de2aa236386ff32f0eef3a0859352e3e2fd","abstract_canon_sha256":"8b3d83310d77d8f55476ce9d02dcd877fb1fecb4d89f195a671c80e299720090"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:46:23.523922Z","signature_b64":"3Lfsc/LOFmVoJGzI9E1E80TsGTF4k4niPaOJ3Ws1spj/CUj8hdaYHqrEal6Kg89o9ZizAL/RPpQYiPKUYQ1gBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"395747d41382aa7705b2e1259b2c31aa7270eb7e0b823987aa391d8a129532b4","last_reissued_at":"2026-07-05T10:46:23.523451Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:46:23.523451Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ShieldGemma 2: Robust and Tractable Image Content Moderation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","eess.IV"],"primary_cat":"cs.CV","authors_text":"Aparna Joshi, Cai Xu, Dana Kurniawan, Dirichi Ike-Njoku, Hamid Palangi, Jindong Gu, Jingjing Zhou, Joon Baek, Karthik Narasimhan, Mani Malek, Rick Pereira, Ryan Mullins, Shravan Dheep, Tamoghna Saha, Wenjun Zeng, Yiwen Song, Yuchi Liu","submitted_at":"2025-04-01T18:00:20Z","abstract_excerpt":"We introduce ShieldGemma 2, a 4B parameter image content moderation model built on Gemma 3. This model provides robust safety risk predictions across the following key harm categories: Sexually Explicit, Violence \\& Gore, and Dangerous Content for synthetic images (e.g. output of any image generation model) and natural images (e.g. any image input to a Vision-Language Model). We evaluated on both internal and external benchmarks to demonstrate state-of-the-art performance compared to LlavaGuard \\citep{helff2024llavaguard}, GPT-4o mini \\citep{hurst2024gpt}, and the base Gemma 3 model \\citep{gem"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.01081","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/2504.01081/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":"2504.01081","created_at":"2026-07-05T10:46:23.523504+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.01081v2","created_at":"2026-07-05T10:46:23.523504+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.01081","created_at":"2026-07-05T10:46:23.523504+00:00"},{"alias_kind":"pith_short_12","alias_value":"HFLUPVATQKVH","created_at":"2026-07-05T10:46:23.523504+00:00"},{"alias_kind":"pith_short_16","alias_value":"HFLUPVATQKVHOBNS","created_at":"2026-07-05T10:46:23.523504+00:00"},{"alias_kind":"pith_short_8","alias_value":"HFLUPVAT","created_at":"2026-07-05T10:46:23.523504+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26199","citing_title":"MIRAGE: Protecting against Malicious Image Editing via False Moderation","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29887","citing_title":"SafePyramid: A Hierarchical Benchmark for In-context Policy Guardrailing","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25893","citing_title":"$D^2$-Monitor: Dynamic Safety Monitoring for Diffusion LLMs via Hesitation-Aware Routing","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26199","citing_title":"MIRAGE: Protecting against Malicious Image Editing via False Moderation","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28137","citing_title":"No Safe Dose: How Training Data Drives Unsafe Image Generation","ref_index":67,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22373","citing_title":"Boundary-targeted Membership Inference Attacks on Safety Classifiers","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22373","citing_title":"Boundary-targeted Membership Inference Attacks on Safety Classifiers","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17610","citing_title":"SafeLens: Deliberate and Efficient Video Guardrails with Fast-and-Slow Screening","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19190","citing_title":"Going PLACES: Participatory Localized Red Teaming for Text-to-Image Safety in the Global South","ref_index":97,"is_internal_anchor":false},{"citing_arxiv_id":"2509.26238","citing_title":"Beyond Linear Probes: Dynamic Safety Monitoring for Language Models","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08819","citing_title":"SenBen: Sensitive Scene Graphs for Explainable Content Moderation","ref_index":40,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ","json":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ.json","graph_json":"https://pith.science/api/pith-number/HFLUPVATQKVHOBNS4ESZWLBRVJ/graph.json","events_json":"https://pith.science/api/pith-number/HFLUPVATQKVHOBNS4ESZWLBRVJ/events.json","paper":"https://pith.science/paper/HFLUPVAT"},"agent_actions":{"view_html":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ","download_json":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ.json","view_paper":"https://pith.science/paper/HFLUPVAT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.01081&json=true","fetch_graph":"https://pith.science/api/pith-number/HFLUPVATQKVHOBNS4ESZWLBRVJ/graph.json","fetch_events":"https://pith.science/api/pith-number/HFLUPVATQKVHOBNS4ESZWLBRVJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ/action/storage_attestation","attest_author":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ/action/author_attestation","sign_citation":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ/action/citation_signature","submit_replication":"https://pith.science/pith/HFLUPVATQKVHOBNS4ESZWLBRVJ/action/replication_record"}},"created_at":"2026-07-05T10:46:23.523504+00:00","updated_at":"2026-07-05T10:46:23.523504+00:00"}