{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:ZBRLIKUZJ4YECWMDURIQJWSFLP","short_pith_number":"pith:ZBRLIKUZ","schema_version":"1.0","canonical_sha256":"c862b42a994f30415983a45104da455bed5f969181c3bccb74002772aa89684f","source":{"kind":"arxiv","id":"2306.13213","version":2},"attestation_state":"computed","paper":{"title":"Visual Adversarial Examples Jailbreak Aligned Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Ashwinee Panda, Kaixuan Huang, Mengdi Wang, Peter Henderson, Prateek Mittal, Xiangyu Qi","submitted_at":"2023-06-22T22:13:03Z","abstract_excerpt":"Recently, there has been a surge of interest in integrating vision into Large Language Models (LLMs), exemplified by Visual Language Models (VLMs) such as Flamingo and GPT-4. This paper sheds light on the security and safety implications of this trend. First, we underscore that the continuous and high-dimensional nature of the visual input makes it a weak link against adversarial attacks, representing an expanded attack surface of vision-integrated LLMs. Second, we highlight that the versatility of LLMs also presents visual attackers with a wider array of achievable adversarial objectives, ext"},"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":"2306.13213","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2023-06-22T22:13:03Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"e2701f7351bb8e0143d100638f7263fcddf33db010ce4cb3b5e1ecd80d53f080","abstract_canon_sha256":"6638149818a3ffec85a7071871d7e31a0f1029e4ec3db7098b8272c80beb9c78"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:42:06.522725Z","signature_b64":"RkFbdpdgUC6aOBlxcmCf3Av+/+cFwv8Sribq8I2A8XN5UIIJ7vZsiSlnOQGzoXPXH3zesGlTQ2Fe9lKky/dHAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c862b42a994f30415983a45104da455bed5f969181c3bccb74002772aa89684f","last_reissued_at":"2026-07-05T06:42:06.522207Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:42:06.522207Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Visual Adversarial Examples Jailbreak Aligned Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Ashwinee Panda, Kaixuan Huang, Mengdi Wang, Peter Henderson, Prateek Mittal, Xiangyu Qi","submitted_at":"2023-06-22T22:13:03Z","abstract_excerpt":"Recently, there has been a surge of interest in integrating vision into Large Language Models (LLMs), exemplified by Visual Language Models (VLMs) such as Flamingo and GPT-4. This paper sheds light on the security and safety implications of this trend. First, we underscore that the continuous and high-dimensional nature of the visual input makes it a weak link against adversarial attacks, representing an expanded attack surface of vision-integrated LLMs. Second, we highlight that the versatility of LLMs also presents visual attackers with a wider array of achievable adversarial objectives, ext"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.13213","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/2306.13213/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":"2306.13213","created_at":"2026-07-05T06:42:06.522272+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.13213v2","created_at":"2026-07-05T06:42:06.522272+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.13213","created_at":"2026-07-05T06:42:06.522272+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZBRLIKUZJ4YE","created_at":"2026-07-05T06:42:06.522272+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZBRLIKUZJ4YECWMD","created_at":"2026-07-05T06:42:06.522272+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZBRLIKUZ","created_at":"2026-07-05T06:42:06.522272+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.20717","citing_title":"MIRAGE: Stealthy Visual Prompt Injection for Vulnerability Detection in Web Agents","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11854","citing_title":"Fine-tuning Multi-modal LLMs with ART: Art-based Reinforcement Training","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09125","citing_title":"Unveiling Privacy Risks in Multi-modal Large Language Models: Task-specific Vulnerabilities and Mitigation Challenges","ref_index":100,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07595","citing_title":"VisualLeakBench: Reproducible Action-Boundary Propagation Failures in Vision-Language Agents","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26566","citing_title":"Adversarial Diffusion Across Modalities: A Fusion Survey of Attacks, Defenses, and Evaluation for Text, Vision, and Vision-Language Models","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17341","citing_title":"Single-Sample Black-Box Membership Inference Attack against Vision-Language Models via Cross-modal Semantic Alignment","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2512.00520","citing_title":"Toward a Safe Internet of Agents","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01758","citing_title":"Catching the Infection Before It Spreads: Foresight-Guided Defense in Multi-Agent Systems","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01449","citing_title":"VisInject: Disruption != Injection -- A Dual-Dimension Evaluation of Universal Adversarial Attacks on Vision-Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2502.18864","citing_title":"Towards an AI co-scientist","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06247","citing_title":"SALLIE: Safeguarding Against Latent Language & Image Exploits","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13803","citing_title":"Gaslight, Gatekeep, V1-V3: Early Visual Cortex Alignment Shields Vision-Language Models from Sycophantic Manipulation","ref_index":42,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP","json":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP.json","graph_json":"https://pith.science/api/pith-number/ZBRLIKUZJ4YECWMDURIQJWSFLP/graph.json","events_json":"https://pith.science/api/pith-number/ZBRLIKUZJ4YECWMDURIQJWSFLP/events.json","paper":"https://pith.science/paper/ZBRLIKUZ"},"agent_actions":{"view_html":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP","download_json":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP.json","view_paper":"https://pith.science/paper/ZBRLIKUZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.13213&json=true","fetch_graph":"https://pith.science/api/pith-number/ZBRLIKUZJ4YECWMDURIQJWSFLP/graph.json","fetch_events":"https://pith.science/api/pith-number/ZBRLIKUZJ4YECWMDURIQJWSFLP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP/action/storage_attestation","attest_author":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP/action/author_attestation","sign_citation":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP/action/citation_signature","submit_replication":"https://pith.science/pith/ZBRLIKUZJ4YECWMDURIQJWSFLP/action/replication_record"}},"created_at":"2026-07-05T06:42:06.522272+00:00","updated_at":"2026-07-05T06:42:06.522272+00:00"}