{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7FNZBWNGQKQTX2R4Z3XULBKX6B","short_pith_number":"pith:7FNZBWNG","schema_version":"1.0","canonical_sha256":"f95b90d9a682a13bea3cceef458557f062efdaa2758f73e17c239f09d1771ff2","source":{"kind":"arxiv","id":"2503.11750","version":1},"attestation_state":"computed","paper":{"title":"Making Every Step Effective: Jailbreaking Large Vision-Language Models Through Hierarchical KV Equalization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.CV","authors_text":"Bryan Hooi, Jun Liu, Muhao Chen, Shuyang Hao, Yiwei Wang, Yujun Cai, Zi Huang","submitted_at":"2025-03-14T17:57:42Z","abstract_excerpt":"In the realm of large vision-language models (LVLMs), adversarial jailbreak attacks serve as a red-teaming approach to identify safety vulnerabilities of these models and their associated defense mechanisms. However, we identify a critical limitation: not every adversarial optimization step leads to a positive outcome, and indiscriminately accepting optimization results at each step may reduce the overall attack success rate. To address this challenge, we introduce HKVE (Hierarchical Key-Value Equalization), an innovative jailbreaking framework that selectively accepts gradient optimization re"},"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":"2503.11750","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-03-14T17:57:42Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"186fdbc425ed1f5c022c446b905607b8a317cc291af782dc3a217c2846011e16","abstract_canon_sha256":"a384a85aca01733dbbd829c0ebe0d466ac53bc237dbe692af1d80f03b88cc4a0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:31:59.233256Z","signature_b64":"2kuDJdZ/ZMrPlVERNzb3aNF9D1VMErt+HqEP22OI2F6bZN/HEIjbTzQ6lOq604Sp7SDAQbRfxgbJx/rIgwaDCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f95b90d9a682a13bea3cceef458557f062efdaa2758f73e17c239f09d1771ff2","last_reissued_at":"2026-07-05T10:31:59.232864Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:31:59.232864Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Making Every Step Effective: Jailbreaking Large Vision-Language Models Through Hierarchical KV Equalization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.CV","authors_text":"Bryan Hooi, Jun Liu, Muhao Chen, Shuyang Hao, Yiwei Wang, Yujun Cai, Zi Huang","submitted_at":"2025-03-14T17:57:42Z","abstract_excerpt":"In the realm of large vision-language models (LVLMs), adversarial jailbreak attacks serve as a red-teaming approach to identify safety vulnerabilities of these models and their associated defense mechanisms. However, we identify a critical limitation: not every adversarial optimization step leads to a positive outcome, and indiscriminately accepting optimization results at each step may reduce the overall attack success rate. To address this challenge, we introduce HKVE (Hierarchical Key-Value Equalization), an innovative jailbreaking framework that selectively accepts gradient optimization re"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.11750","kind":"arxiv","version":1},"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/2503.11750/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":"2503.11750","created_at":"2026-07-05T10:31:59.232920+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.11750v1","created_at":"2026-07-05T10:31:59.232920+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.11750","created_at":"2026-07-05T10:31:59.232920+00:00"},{"alias_kind":"pith_short_12","alias_value":"7FNZBWNGQKQT","created_at":"2026-07-05T10:31:59.232920+00:00"},{"alias_kind":"pith_short_16","alias_value":"7FNZBWNGQKQTX2R4","created_at":"2026-07-05T10:31:59.232920+00:00"},{"alias_kind":"pith_short_8","alias_value":"7FNZBWNG","created_at":"2026-07-05T10:31:59.232920+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.15030","citing_title":"WARD: Adversarially Robust Defense of Web Agents Against Prompt Injections","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2604.10299","citing_title":"Seeing No Evil: Blinding Large Vision-Language Models to Safety Instructions via Adversarial Attention Hijacking","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B","json":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B.json","graph_json":"https://pith.science/api/pith-number/7FNZBWNGQKQTX2R4Z3XULBKX6B/graph.json","events_json":"https://pith.science/api/pith-number/7FNZBWNGQKQTX2R4Z3XULBKX6B/events.json","paper":"https://pith.science/paper/7FNZBWNG"},"agent_actions":{"view_html":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B","download_json":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B.json","view_paper":"https://pith.science/paper/7FNZBWNG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.11750&json=true","fetch_graph":"https://pith.science/api/pith-number/7FNZBWNGQKQTX2R4Z3XULBKX6B/graph.json","fetch_events":"https://pith.science/api/pith-number/7FNZBWNGQKQTX2R4Z3XULBKX6B/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B/action/storage_attestation","attest_author":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B/action/author_attestation","sign_citation":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B/action/citation_signature","submit_replication":"https://pith.science/pith/7FNZBWNGQKQTX2R4Z3XULBKX6B/action/replication_record"}},"created_at":"2026-07-05T10:31:59.232920+00:00","updated_at":"2026-07-05T10:31:59.232920+00:00"}