{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:3LLUODIW6CU7LFESOX3VFGR3AW","short_pith_number":"pith:3LLUODIW","schema_version":"1.0","canonical_sha256":"dad7470d16f0a9f5949275f7529a3b05bff174a87fea31f0224f42792dbff90d","source":{"kind":"arxiv","id":"2508.09456","version":5},"attestation_state":"computed","paper":{"title":"IAG: Input-aware Backdoor Attack on VLM-based Visual Grounding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.CV","authors_text":"Beining Xu, Di Zhang, Haodong Zhao, Jiatong Li, Jingdi Lei, Junxian Li, Simin Chen","submitted_at":"2025-08-13T03:22:19Z","abstract_excerpt":"Recent advances in vision-language models (VLMs) have significantly enhanced the visual grounding task, which involves locating objects in an image based on natural language queries. Despite these advancements, the security of VLM-based grounding systems has not been thoroughly investigated. This paper reveals a novel and realistic vulnerability: the first multi-target backdoor attack on VLM-based visual grounding. Unlike prior attacks that rely on static triggers or fixed targets, we propose IAG, a method that dynamically generates input-aware, text-guided triggers conditioned on any specifie"},"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":"2508.09456","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-08-13T03:22:19Z","cross_cats_sorted":["cs.CL","cs.CR"],"title_canon_sha256":"222d18adb3c3bdaf4f426f4d49499d3dbe96f7b0ef4ca6741d5d9edd340d23e4","abstract_canon_sha256":"666be436f13041e1e57b0400a1493d35aa4c03294fd667fe1ebecd5ca5bc88c4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T01:03:33.208562Z","signature_b64":"djCO9Yj6D9lLm/m1jAiUj/oGSj+0fDZ5jemYwZfplmWdXXZrzLMbRHptzrpuTnj44os+otVPPW0PT6+m2tyxDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dad7470d16f0a9f5949275f7529a3b05bff174a87fea31f0224f42792dbff90d","last_reissued_at":"2026-06-02T01:03:33.208046Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T01:03:33.208046Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"IAG: Input-aware Backdoor Attack on VLM-based Visual Grounding","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.CV","authors_text":"Beining Xu, Di Zhang, Haodong Zhao, Jiatong Li, Jingdi Lei, Junxian Li, Simin Chen","submitted_at":"2025-08-13T03:22:19Z","abstract_excerpt":"Recent advances in vision-language models (VLMs) have significantly enhanced the visual grounding task, which involves locating objects in an image based on natural language queries. Despite these advancements, the security of VLM-based grounding systems has not been thoroughly investigated. This paper reveals a novel and realistic vulnerability: the first multi-target backdoor attack on VLM-based visual grounding. Unlike prior attacks that rely on static triggers or fixed targets, we propose IAG, a method that dynamically generates input-aware, text-guided triggers conditioned on any specifie"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.09456","kind":"arxiv","version":5},"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/2508.09456/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":"2508.09456","created_at":"2026-06-02T01:03:33.208110+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.09456v5","created_at":"2026-06-02T01:03:33.208110+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.09456","created_at":"2026-06-02T01:03:33.208110+00:00"},{"alias_kind":"pith_short_12","alias_value":"3LLUODIW6CU7","created_at":"2026-06-02T01:03:33.208110+00:00"},{"alias_kind":"pith_short_16","alias_value":"3LLUODIW6CU7LFES","created_at":"2026-06-02T01:03:33.208110+00:00"},{"alias_kind":"pith_short_8","alias_value":"3LLUODIW","created_at":"2026-06-02T01:03:33.208110+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":4,"sample":[{"citing_arxiv_id":"2605.18603","citing_title":"Starve to Perceive: Taming Lazy Perception in VLMs with Constrained Visual Bandwidth","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2604.24348","citing_title":"OS-SPEAR: A Toolkit for the Safety, Performance,Efficiency, and Robustness Analysis of OS Agents","ref_index":86,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19083","citing_title":"ProjLens: Unveiling the Role of Projectors in Multimodal Model Safety","ref_index":146,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08395","citing_title":"Phantasia: Context-Adaptive Backdoors in Vision Language Models","ref_index":13,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW","json":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW.json","graph_json":"https://pith.science/api/pith-number/3LLUODIW6CU7LFESOX3VFGR3AW/graph.json","events_json":"https://pith.science/api/pith-number/3LLUODIW6CU7LFESOX3VFGR3AW/events.json","paper":"https://pith.science/paper/3LLUODIW"},"agent_actions":{"view_html":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW","download_json":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW.json","view_paper":"https://pith.science/paper/3LLUODIW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.09456&json=true","fetch_graph":"https://pith.science/api/pith-number/3LLUODIW6CU7LFESOX3VFGR3AW/graph.json","fetch_events":"https://pith.science/api/pith-number/3LLUODIW6CU7LFESOX3VFGR3AW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW/action/storage_attestation","attest_author":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW/action/author_attestation","sign_citation":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW/action/citation_signature","submit_replication":"https://pith.science/pith/3LLUODIW6CU7LFESOX3VFGR3AW/action/replication_record"}},"created_at":"2026-06-02T01:03:33.208110+00:00","updated_at":"2026-06-02T01:03:33.208110+00:00"}