{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NXI6KLT66EEUM3RVSNJQRSN5PJ","short_pith_number":"pith:NXI6KLT6","schema_version":"1.0","canonical_sha256":"6dd1e52e7ef109466e35935308c9bd7a55351ff59dffaf6cec92f3840d36b37c","source":{"kind":"arxiv","id":"2407.09050","version":2},"attestation_state":"computed","paper":{"title":"Refusing Safe Prompts for Multi-modal Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.LG"],"primary_cat":"cs.CR","authors_text":"Hongbin Liu, Neil Zhenqiang Gong, Yuepeng Hu, Zedian Shao","submitted_at":"2024-07-12T07:18:05Z","abstract_excerpt":"Multimodal large language models (MLLMs) have become the cornerstone of today's generative AI ecosystem, sparking intense competition among tech giants and startups. In particular, an MLLM generates a text response given a prompt consisting of an image and a question. While state-of-the-art MLLMs use safety filters and alignment techniques to refuse unsafe prompts, in this work, we introduce MLLM-Refusal, the first method that induces refusals for safe prompts. In particular, our MLLM-Refusal optimizes a nearly-imperceptible refusal perturbation and adds it to an image, causing target MLLMs to"},"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":"2407.09050","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2024-07-12T07:18:05Z","cross_cats_sorted":["cs.AI","cs.CV","cs.LG"],"title_canon_sha256":"a2886c8b8d8d904e9a121d15200b511138694a8486d372dc170ac5e859529e2c","abstract_canon_sha256":"6a0b412d8e869730fb1bd40839a90b94c5b434b6e8b774f14ca962f13fad2700"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:03:54.546547Z","signature_b64":"2M7wRT3fRgncq/x9gKXeAvsCnVdubkbcjX+9s9smVSnD0GvAbhr6S1qdt9xCi3Jt7HR7OeTGCJvRA6T98qS2CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6dd1e52e7ef109466e35935308c9bd7a55351ff59dffaf6cec92f3840d36b37c","last_reissued_at":"2026-07-05T09:03:54.546099Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:03:54.546099Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Refusing Safe Prompts for Multi-modal Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.LG"],"primary_cat":"cs.CR","authors_text":"Hongbin Liu, Neil Zhenqiang Gong, Yuepeng Hu, Zedian Shao","submitted_at":"2024-07-12T07:18:05Z","abstract_excerpt":"Multimodal large language models (MLLMs) have become the cornerstone of today's generative AI ecosystem, sparking intense competition among tech giants and startups. In particular, an MLLM generates a text response given a prompt consisting of an image and a question. While state-of-the-art MLLMs use safety filters and alignment techniques to refuse unsafe prompts, in this work, we introduce MLLM-Refusal, the first method that induces refusals for safe prompts. In particular, our MLLM-Refusal optimizes a nearly-imperceptible refusal perturbation and adds it to an image, causing target MLLMs to"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.09050","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/2407.09050/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":"2407.09050","created_at":"2026-07-05T09:03:54.546158+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.09050v2","created_at":"2026-07-05T09:03:54.546158+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.09050","created_at":"2026-07-05T09:03:54.546158+00:00"},{"alias_kind":"pith_short_12","alias_value":"NXI6KLT66EEU","created_at":"2026-07-05T09:03:54.546158+00:00"},{"alias_kind":"pith_short_16","alias_value":"NXI6KLT66EEUM3RV","created_at":"2026-07-05T09:03:54.546158+00:00"},{"alias_kind":"pith_short_8","alias_value":"NXI6KLT6","created_at":"2026-07-05T09:03:54.546158+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21147","citing_title":"AOR-Bench: Do Large Audio Language Models Over-Refuse Pseudo-Harmful Queries?","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2508.11222","citing_title":"ORFuzz: Fuzzing the \"Other Side\" of LLM Safety -- Testing Over-Refusal","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ","json":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ.json","graph_json":"https://pith.science/api/pith-number/NXI6KLT66EEUM3RVSNJQRSN5PJ/graph.json","events_json":"https://pith.science/api/pith-number/NXI6KLT66EEUM3RVSNJQRSN5PJ/events.json","paper":"https://pith.science/paper/NXI6KLT6"},"agent_actions":{"view_html":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ","download_json":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ.json","view_paper":"https://pith.science/paper/NXI6KLT6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.09050&json=true","fetch_graph":"https://pith.science/api/pith-number/NXI6KLT66EEUM3RVSNJQRSN5PJ/graph.json","fetch_events":"https://pith.science/api/pith-number/NXI6KLT66EEUM3RVSNJQRSN5PJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ/action/storage_attestation","attest_author":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ/action/author_attestation","sign_citation":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ/action/citation_signature","submit_replication":"https://pith.science/pith/NXI6KLT66EEUM3RVSNJQRSN5PJ/action/replication_record"}},"created_at":"2026-07-05T09:03:54.546158+00:00","updated_at":"2026-07-05T09:03:54.546158+00:00"}