{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:7UPFYHKUACJLSOOBHW5IPFRVGE","short_pith_number":"pith:7UPFYHKU","schema_version":"1.0","canonical_sha256":"fd1e5c1d540092b939c13dba87963531085c37124bfc38a8cc0a5e61cc924b93","source":{"kind":"arxiv","id":"2411.01703","version":2},"attestation_state":"computed","paper":{"title":"UniGuard: Towards Universal Safety Guardrails for Jailbreak Attacks on Multimodal Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Donghyun Kim, Eric Ma, Gaurav Verma, Megha Sharma, Sejoon Oh, Srijan Kumar, Yiqiao Jin","submitted_at":"2024-11-03T22:19:20Z","abstract_excerpt":"Multimodal large language models (MLLMs) have revolutionized vision-language understanding but remain vulnerable to multimodal jailbreak attacks, where adversarial inputs are meticulously crafted to elicit harmful or inappropriate responses. We propose UniGuard, a novel multimodal safety guardrail that jointly considers the unimodal and cross-modal harmful signals. UniGuard trains a multimodal guardrail to minimize the likelihood of generating harmful responses in a toxic corpus. The guardrail can be seamlessly applied to any input prompt during inference with minimal computational costs. Exte"},"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":"2411.01703","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-11-03T22:19:20Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"25ec94b2f976ff6498fdbd165f81dc4275c66ca4c8b4dc193488ce95e6dff13b","abstract_canon_sha256":"bf5a2d959314d362b425c23ce51d55cd86a38f15d9e3329f3c16116157cf22ec"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:07:45.882509Z","signature_b64":"DXVrP+RSQjc2S199C03/WBKFSHdvpV7ESv5qeXh23EfdHQqz8CPqqscIECxD3f0Nh+1t24wIV/vipxixCTmADg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fd1e5c1d540092b939c13dba87963531085c37124bfc38a8cc0a5e61cc924b93","last_reissued_at":"2026-07-05T10:07:45.882060Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:07:45.882060Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"UniGuard: Towards Universal Safety Guardrails for Jailbreak Attacks on Multimodal Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Donghyun Kim, Eric Ma, Gaurav Verma, Megha Sharma, Sejoon Oh, Srijan Kumar, Yiqiao Jin","submitted_at":"2024-11-03T22:19:20Z","abstract_excerpt":"Multimodal large language models (MLLMs) have revolutionized vision-language understanding but remain vulnerable to multimodal jailbreak attacks, where adversarial inputs are meticulously crafted to elicit harmful or inappropriate responses. We propose UniGuard, a novel multimodal safety guardrail that jointly considers the unimodal and cross-modal harmful signals. UniGuard trains a multimodal guardrail to minimize the likelihood of generating harmful responses in a toxic corpus. The guardrail can be seamlessly applied to any input prompt during inference with minimal computational costs. Exte"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.01703","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/2411.01703/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":"2411.01703","created_at":"2026-07-05T10:07:45.882118+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.01703v2","created_at":"2026-07-05T10:07:45.882118+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.01703","created_at":"2026-07-05T10:07:45.882118+00:00"},{"alias_kind":"pith_short_12","alias_value":"7UPFYHKUACJL","created_at":"2026-07-05T10:07:45.882118+00:00"},{"alias_kind":"pith_short_16","alias_value":"7UPFYHKUACJLSOOB","created_at":"2026-07-05T10:07:45.882118+00:00"},{"alias_kind":"pith_short_8","alias_value":"7UPFYHKU","created_at":"2026-07-05T10:07:45.882118+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2503.21460","citing_title":"Large Language Model Agent: A Survey on Methodology, Applications and Challenges","ref_index":251,"is_internal_anchor":false},{"citing_arxiv_id":"2604.25562","citing_title":"SnapGuard: Lightweight Prompt Injection Detection for Screenshot-Based Web Agents","ref_index":32,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE","json":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE.json","graph_json":"https://pith.science/api/pith-number/7UPFYHKUACJLSOOBHW5IPFRVGE/graph.json","events_json":"https://pith.science/api/pith-number/7UPFYHKUACJLSOOBHW5IPFRVGE/events.json","paper":"https://pith.science/paper/7UPFYHKU"},"agent_actions":{"view_html":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE","download_json":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE.json","view_paper":"https://pith.science/paper/7UPFYHKU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.01703&json=true","fetch_graph":"https://pith.science/api/pith-number/7UPFYHKUACJLSOOBHW5IPFRVGE/graph.json","fetch_events":"https://pith.science/api/pith-number/7UPFYHKUACJLSOOBHW5IPFRVGE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE/action/storage_attestation","attest_author":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE/action/author_attestation","sign_citation":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE/action/citation_signature","submit_replication":"https://pith.science/pith/7UPFYHKUACJLSOOBHW5IPFRVGE/action/replication_record"}},"created_at":"2026-07-05T10:07:45.882118+00:00","updated_at":"2026-07-05T10:07:45.882118+00:00"}