{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:SYNRGLEUYOHKG3ZP5A43KX5SVK","short_pith_number":"pith:SYNRGLEU","schema_version":"1.0","canonical_sha256":"961b132c94c38ea36f2fe839b55fb2aab088ae5c4126c16b26284ba1ecf8672b","source":{"kind":"arxiv","id":"2411.18000","version":2},"attestation_state":"computed","paper":{"title":"Exploring Visual Vulnerabilities via Multi-Loss Adversarial Search for Jailbreaking Vision-Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bryan Hooi, Jun Liu, Kai-Wei Chang, Shuyang Hao, Yujun Cai, Zi Huang","submitted_at":"2024-11-27T02:40:29Z","abstract_excerpt":"Despite inheriting security measures from underlying language models, Vision-Language Models (VLMs) may still be vulnerable to safety alignment issues. Through empirical analysis, we uncover two critical findings: scenario-matched images can significantly amplify harmful outputs, and contrary to common assumptions in gradient-based attacks, minimal loss values do not guarantee optimal attack effectiveness. Building on these insights, we introduce MLAI (Multi-Loss Adversarial Images), a novel jailbreak framework that leverages scenario-aware image generation for semantic alignment, exploits fla"},"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":"2411.18000","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2024-11-27T02:40:29Z","cross_cats_sorted":[],"title_canon_sha256":"67666a5a384f99aec48dcd39c3f3c10135513f9d12144330475adef92069cd48","abstract_canon_sha256":"f2a610c6153ad0f895b47e6702b48869816643342ffd76a72e063c8bb59938d3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:41:47.240203Z","signature_b64":"Y+EG9cxwiCByxRkLJMn89HOJMpcFvhl8FHxGaCOCf8BYdWbEEsN/ExmytcsSmUE3annP+jxLcs6sFCtavo3ABA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"961b132c94c38ea36f2fe839b55fb2aab088ae5c4126c16b26284ba1ecf8672b","last_reissued_at":"2026-07-05T09:41:47.239755Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:41:47.239755Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Exploring Visual Vulnerabilities via Multi-Loss Adversarial Search for Jailbreaking Vision-Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bryan Hooi, Jun Liu, Kai-Wei Chang, Shuyang Hao, Yujun Cai, Zi Huang","submitted_at":"2024-11-27T02:40:29Z","abstract_excerpt":"Despite inheriting security measures from underlying language models, Vision-Language Models (VLMs) may still be vulnerable to safety alignment issues. Through empirical analysis, we uncover two critical findings: scenario-matched images can significantly amplify harmful outputs, and contrary to common assumptions in gradient-based attacks, minimal loss values do not guarantee optimal attack effectiveness. Building on these insights, we introduce MLAI (Multi-Loss Adversarial Images), a novel jailbreak framework that leverages scenario-aware image generation for semantic alignment, exploits fla"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.18000","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/2411.18000/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":"2411.18000","created_at":"2026-07-05T09:41:47.239809+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.18000v2","created_at":"2026-07-05T09:41:47.239809+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.18000","created_at":"2026-07-05T09:41:47.239809+00:00"},{"alias_kind":"pith_short_12","alias_value":"SYNRGLEUYOHK","created_at":"2026-07-05T09:41:47.239809+00:00"},{"alias_kind":"pith_short_16","alias_value":"SYNRGLEUYOHKG3ZP","created_at":"2026-07-05T09:41:47.239809+00:00"},{"alias_kind":"pith_short_8","alias_value":"SYNRGLEU","created_at":"2026-07-05T09:41:47.239809+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.11968","citing_title":"Watch, Listen, Understand, Mislead: Tri-modal Adversarial Attacks on Short Videos for Content Appropriateness Evaluation","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK","json":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK.json","graph_json":"https://pith.science/api/pith-number/SYNRGLEUYOHKG3ZP5A43KX5SVK/graph.json","events_json":"https://pith.science/api/pith-number/SYNRGLEUYOHKG3ZP5A43KX5SVK/events.json","paper":"https://pith.science/paper/SYNRGLEU"},"agent_actions":{"view_html":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK","download_json":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK.json","view_paper":"https://pith.science/paper/SYNRGLEU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.18000&json=true","fetch_graph":"https://pith.science/api/pith-number/SYNRGLEUYOHKG3ZP5A43KX5SVK/graph.json","fetch_events":"https://pith.science/api/pith-number/SYNRGLEUYOHKG3ZP5A43KX5SVK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK/action/storage_attestation","attest_author":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK/action/author_attestation","sign_citation":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK/action/citation_signature","submit_replication":"https://pith.science/pith/SYNRGLEUYOHKG3ZP5A43KX5SVK/action/replication_record"}},"created_at":"2026-07-05T09:41:47.239809+00:00","updated_at":"2026-07-05T09:41:47.239809+00:00"}