{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:5WG3ILU7H4CZFNM4JYWLT7UXHJ","short_pith_number":"pith:5WG3ILU7","schema_version":"1.0","canonical_sha256":"ed8db42e9f3f0592b59c4e2cb9fe973a5a1e89ae7a340bf62793c5ad0625c433","source":{"kind":"arxiv","id":"2402.02309","version":1},"attestation_state":"computed","paper":{"title":"Jailbreaking Attack against Multimodal Large Language Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR","cs.CV"],"primary_cat":"cs.LG","authors_text":"Gang Hua, Haodong Ren, Rong Jin, Xinbo Gao, Zhenxing Niu","submitted_at":"2024-02-04T01:29:24Z","abstract_excerpt":"This paper focuses on jailbreaking attacks against multi-modal large language models (MLLMs), seeking to elicit MLLMs to generate objectionable responses to harmful user queries. A maximum likelihood-based algorithm is proposed to find an \\emph{image Jailbreaking Prompt} (imgJP), enabling jailbreaks against MLLMs across multiple unseen prompts and images (i.e., data-universal property). Our approach exhibits strong model-transferability, as the generated imgJP can be transferred to jailbreak various models, including MiniGPT-v2, LLaVA, InstructBLIP, and mPLUG-Owl2, in a black-box manner. Moreo"},"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":"2402.02309","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-04T01:29:24Z","cross_cats_sorted":["cs.CL","cs.CR","cs.CV"],"title_canon_sha256":"91f1bb049ea3a4cfa25c4aed9807607b2c0f3d295cad69c9d9af98ee3702ae86","abstract_canon_sha256":"d883a385de079e51c3b41497fd38756466a488a07dafc316ea76f3ed22f20f8e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:41:48.454982Z","signature_b64":"7SpwCPhGHUvqzZX6wzVRQp7MGD45ijbkiQRe/v9iTLWO4e5t36jkbOYvkcXnO6ModbNDNLfThpjkeFZ98l9TBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ed8db42e9f3f0592b59c4e2cb9fe973a5a1e89ae7a340bf62793c5ad0625c433","last_reissued_at":"2026-07-05T07:41:48.454567Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:41:48.454567Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Jailbreaking Attack against Multimodal Large Language Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR","cs.CV"],"primary_cat":"cs.LG","authors_text":"Gang Hua, Haodong Ren, Rong Jin, Xinbo Gao, Zhenxing Niu","submitted_at":"2024-02-04T01:29:24Z","abstract_excerpt":"This paper focuses on jailbreaking attacks against multi-modal large language models (MLLMs), seeking to elicit MLLMs to generate objectionable responses to harmful user queries. A maximum likelihood-based algorithm is proposed to find an \\emph{image Jailbreaking Prompt} (imgJP), enabling jailbreaks against MLLMs across multiple unseen prompts and images (i.e., data-universal property). Our approach exhibits strong model-transferability, as the generated imgJP can be transferred to jailbreak various models, including MiniGPT-v2, LLaVA, InstructBLIP, and mPLUG-Owl2, in a black-box manner. Moreo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.02309","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/2402.02309/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":"2402.02309","created_at":"2026-07-05T07:41:48.454616+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.02309v1","created_at":"2026-07-05T07:41:48.454616+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.02309","created_at":"2026-07-05T07:41:48.454616+00:00"},{"alias_kind":"pith_short_12","alias_value":"5WG3ILU7H4CZ","created_at":"2026-07-05T07:41:48.454616+00:00"},{"alias_kind":"pith_short_16","alias_value":"5WG3ILU7H4CZFNM4","created_at":"2026-07-05T07:41:48.454616+00:00"},{"alias_kind":"pith_short_8","alias_value":"5WG3ILU7","created_at":"2026-07-05T07:41:48.454616+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24388","citing_title":"PHANTOM: A Large-Scale Dataset of Multimodal Adversarial Attacks for Vision-Language Models","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2606.10904","citing_title":"When the Defense Writes the Refusal: Auditing Keyword-Scored Evaluation of Inference-Time Defenses for Multimodal Large Language Models","ref_index":18,"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":85,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03713","citing_title":"Investigating Adversarial Robustness of Multi-modal Large Language Models","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23157","citing_title":"Same Model, Different Weakness: How Language and Modality Reshape the Jailbreak Attack Surface in Frontier MLLMs","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2502.05206","citing_title":"Safety at Scale: A Comprehensive Survey of Large Model and Agent Safety","ref_index":277,"is_internal_anchor":false},{"citing_arxiv_id":"2503.06223","citing_title":"RedDiffuser: Auditing Multimodal Safety Failures in Vision-Language Models via Reinforced Diffusion","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17248","citing_title":"Image-to-Video Diffusion: From Foundations to Open Frontiers","ref_index":190,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17310","citing_title":"Attention Hijacking: Response Manipulation Across Queries in Vision-Language Models","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11716","citing_title":"SafeSteer: A Decoding-level Defense Mechanism for Multimodal Large Language Models","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27818","citing_title":"MASCing: Configurable Mixture-of-Experts Behavior via Activation Steering Masks","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05058","citing_title":"SoK: Robustness in Large Language Models against Jailbreak Attacks","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02946","citing_title":"RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMs","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12616","citing_title":"Every Picture Tells a Dangerous Story: Memory-Augmented Multi-Agent Jailbreak Attacks on VLMs","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12232","citing_title":"TEMPLATEFUZZ: Fine-Grained Chat Template Fuzzing for Jailbreaking and Red Teaming LLMs","ref_index":37,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ","json":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ.json","graph_json":"https://pith.science/api/pith-number/5WG3ILU7H4CZFNM4JYWLT7UXHJ/graph.json","events_json":"https://pith.science/api/pith-number/5WG3ILU7H4CZFNM4JYWLT7UXHJ/events.json","paper":"https://pith.science/paper/5WG3ILU7"},"agent_actions":{"view_html":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ","download_json":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ.json","view_paper":"https://pith.science/paper/5WG3ILU7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.02309&json=true","fetch_graph":"https://pith.science/api/pith-number/5WG3ILU7H4CZFNM4JYWLT7UXHJ/graph.json","fetch_events":"https://pith.science/api/pith-number/5WG3ILU7H4CZFNM4JYWLT7UXHJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ/action/storage_attestation","attest_author":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ/action/author_attestation","sign_citation":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ/action/citation_signature","submit_replication":"https://pith.science/pith/5WG3ILU7H4CZFNM4JYWLT7UXHJ/action/replication_record"}},"created_at":"2026-07-05T07:41:48.454616+00:00","updated_at":"2026-07-05T07:41:48.454616+00:00"}