{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:OCXU5552U6UCIWNICCIDO3P4FL","short_pith_number":"pith:OCXU5552","schema_version":"1.0","canonical_sha256":"70af4ef7baa7a82459a81090376dfc2aff967cf7117f0ab7896a48a94b1e0c76","source":{"kind":"arxiv","id":"2403.09513","version":1},"attestation_state":"computed","paper":{"title":"AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Chaowei Xiao, Muhao Chen, Xiaogeng Liu, Yu Li, Yu Wang","submitted_at":"2024-03-14T15:57:13Z","abstract_excerpt":"With the advent and widespread deployment of Multimodal Large Language Models (MLLMs), the imperative to ensure their safety has become increasingly pronounced. However, with the integration of additional modalities, MLLMs are exposed to new vulnerabilities, rendering them prone to structured-based jailbreak attacks, where semantic content (e.g., \"harmful text\") has been injected into the images to mislead MLLMs. In this work, we aim to defend against such threats. Specifically, we propose \\textbf{Ada}ptive \\textbf{Shield} Prompting (\\textbf{AdaShield}), which prepends inputs with defense prom"},"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":"2403.09513","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CR","submitted_at":"2024-03-14T15:57:13Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"214ad1e81eeb249a4da067c949e06c2b2a0e9379c64134ddcc31e9bd5a03408f","abstract_canon_sha256":"3c444dc29ab078571dc1f41c21d9acb9f2dd616ad8aa68f83c6c31e829b8dc54"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:56:10.927748Z","signature_b64":"T0dAwAy/jrjLdUvn3x/cmASr9Tl83afn6Mw57Lnalwi+J7UEUL7XfCaIuvc06nxMymbzj+/ia93EWnhBup1iCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"70af4ef7baa7a82459a81090376dfc2aff967cf7117f0ab7896a48a94b1e0c76","last_reissued_at":"2026-07-05T07:56:10.927144Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:56:10.927144Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AdaShield: Safeguarding Multimodal Large Language Models from Structure-based Attack via Adaptive Shield Prompting","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Chaowei Xiao, Muhao Chen, Xiaogeng Liu, Yu Li, Yu Wang","submitted_at":"2024-03-14T15:57:13Z","abstract_excerpt":"With the advent and widespread deployment of Multimodal Large Language Models (MLLMs), the imperative to ensure their safety has become increasingly pronounced. However, with the integration of additional modalities, MLLMs are exposed to new vulnerabilities, rendering them prone to structured-based jailbreak attacks, where semantic content (e.g., \"harmful text\") has been injected into the images to mislead MLLMs. In this work, we aim to defend against such threats. Specifically, we propose \\textbf{Ada}ptive \\textbf{Shield} Prompting (\\textbf{AdaShield}), which prepends inputs with defense prom"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.09513","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/2403.09513/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":"2403.09513","created_at":"2026-07-05T07:56:10.927211+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.09513v1","created_at":"2026-07-05T07:56:10.927211+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.09513","created_at":"2026-07-05T07:56:10.927211+00:00"},{"alias_kind":"pith_short_12","alias_value":"OCXU5552U6UC","created_at":"2026-07-05T07:56:10.927211+00:00"},{"alias_kind":"pith_short_16","alias_value":"OCXU5552U6UCIWNI","created_at":"2026-07-05T07:56:10.927211+00:00"},{"alias_kind":"pith_short_8","alias_value":"OCXU5552","created_at":"2026-07-05T07:56:10.927211+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09125","citing_title":"Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges","ref_index":107,"is_internal_anchor":false},{"citing_arxiv_id":"2603.21697","citing_title":"Structured Visual Narratives Undermine Safety Alignment in Multimodal Large Language Models","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL","json":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL.json","graph_json":"https://pith.science/api/pith-number/OCXU5552U6UCIWNICCIDO3P4FL/graph.json","events_json":"https://pith.science/api/pith-number/OCXU5552U6UCIWNICCIDO3P4FL/events.json","paper":"https://pith.science/paper/OCXU5552"},"agent_actions":{"view_html":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL","download_json":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL.json","view_paper":"https://pith.science/paper/OCXU5552","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.09513&json=true","fetch_graph":"https://pith.science/api/pith-number/OCXU5552U6UCIWNICCIDO3P4FL/graph.json","fetch_events":"https://pith.science/api/pith-number/OCXU5552U6UCIWNICCIDO3P4FL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL/action/storage_attestation","attest_author":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL/action/author_attestation","sign_citation":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL/action/citation_signature","submit_replication":"https://pith.science/pith/OCXU5552U6UCIWNICCIDO3P4FL/action/replication_record"}},"created_at":"2026-07-05T07:56:10.927211+00:00","updated_at":"2026-07-05T07:56:10.927211+00:00"}