{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:J2PB2LJGUNORKN55BTDPJEP3S2","short_pith_number":"pith:J2PB2LJG","schema_version":"1.0","canonical_sha256":"4e9e1d2d26a35d1537bd0cc6f491fb969d99c6df2a895af390bdcb9ddf2f9db4","source":{"kind":"arxiv","id":"2505.21494","version":1},"attestation_state":"computed","paper":{"title":"Adversarial Attacks against Closed-Source MLLMs via Feature Optimal Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bo Li, Chao Du, Sensen Gao, Simeng Qin, Tianyu Pang, Xiaojun Jia, Xinfeng Li, Yang Liu, Yihao Huang, Yiming Li","submitted_at":"2025-05-27T17:56:57Z","abstract_excerpt":"Multimodal large language models (MLLMs) remain vulnerable to transferable adversarial examples. While existing methods typically achieve targeted attacks by aligning global features-such as CLIP's [CLS] token-between adversarial and target samples, they often overlook the rich local information encoded in patch tokens. This leads to suboptimal alignment and limited transferability, particularly for closed-source models. To address this limitation, we propose a targeted transferable adversarial attack method based on feature optimal alignment, called FOA-Attack, to improve adversarial transfer"},"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":"2505.21494","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-05-27T17:56:57Z","cross_cats_sorted":[],"title_canon_sha256":"7bd0fe02f08b10393cd5c2b06f6877da5455a61d8e5c3f99ad42b5481c8a90d5","abstract_canon_sha256":"2bc5334415ce1591034afbd2e75275d5f543b9fc0c672cff4210e7ac75e08567"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:42.802710Z","signature_b64":"5Cj3Sse4X1d93zqJ4BvihCpKWc8zJefuprwsRJiiEgtBPMWbmbYmMydIi0S4rS7x1Limlg5x+BS2fU2Ekd7sAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4e9e1d2d26a35d1537bd0cc6f491fb969d99c6df2a895af390bdcb9ddf2f9db4","last_reissued_at":"2026-07-05T11:10:42.802223Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:42.802223Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Adversarial Attacks against Closed-Source MLLMs via Feature Optimal Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bo Li, Chao Du, Sensen Gao, Simeng Qin, Tianyu Pang, Xiaojun Jia, Xinfeng Li, Yang Liu, Yihao Huang, Yiming Li","submitted_at":"2025-05-27T17:56:57Z","abstract_excerpt":"Multimodal large language models (MLLMs) remain vulnerable to transferable adversarial examples. While existing methods typically achieve targeted attacks by aligning global features-such as CLIP's [CLS] token-between adversarial and target samples, they often overlook the rich local information encoded in patch tokens. This leads to suboptimal alignment and limited transferability, particularly for closed-source models. To address this limitation, we propose a targeted transferable adversarial attack method based on feature optimal alignment, called FOA-Attack, to improve adversarial transfer"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.21494","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/2505.21494/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":"2505.21494","created_at":"2026-07-05T11:10:42.802280+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.21494v1","created_at":"2026-07-05T11:10:42.802280+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.21494","created_at":"2026-07-05T11:10:42.802280+00:00"},{"alias_kind":"pith_short_12","alias_value":"J2PB2LJGUNOR","created_at":"2026-07-05T11:10:42.802280+00:00"},{"alias_kind":"pith_short_16","alias_value":"J2PB2LJGUNORKN55","created_at":"2026-07-05T11:10:42.802280+00:00"},{"alias_kind":"pith_short_8","alias_value":"J2PB2LJG","created_at":"2026-07-05T11:10:42.802280+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07375","citing_title":"On Adversarial Vulnerability of Vision-Language Models through the Lens of Intermediate Spectral Subspaces","ref_index":37,"is_internal_anchor":true},{"citing_arxiv_id":"2606.26199","citing_title":"MIRAGE: Protecting against Malicious Image Editing via False Moderation","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.23892","citing_title":"REALM: A Unified Red-Teaming Benchmark for Physical-World VLMs","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.26199","citing_title":"MIRAGE: Protecting against Malicious Image Editing via False Moderation","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29602","citing_title":"CogniVerse: Revolutionizing Multi-Modal Retrieval-Augmented Generation with Cognitive Reflection and Geometric Reasoning","ref_index":62,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18868","citing_title":"DarkLLM: Learning Language-Driven Adversarial Attacks with Large Language Models","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2601.23179","citing_title":"Universal Adversarial Attacks against Closed-Source MLLMs via Target-View Routed Meta Optimization","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09902","citing_title":"Adversarial Attacks Against MLLMs via Progressive Resolution Processing and Adaptive Feature Alignment","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09253","citing_title":"Mosaic: Multimodal Jailbreak against Closed-Source VLMs via Multi-View Ensemble Optimization","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2","json":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2.json","graph_json":"https://pith.science/api/pith-number/J2PB2LJGUNORKN55BTDPJEP3S2/graph.json","events_json":"https://pith.science/api/pith-number/J2PB2LJGUNORKN55BTDPJEP3S2/events.json","paper":"https://pith.science/paper/J2PB2LJG"},"agent_actions":{"view_html":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2","download_json":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2.json","view_paper":"https://pith.science/paper/J2PB2LJG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.21494&json=true","fetch_graph":"https://pith.science/api/pith-number/J2PB2LJGUNORKN55BTDPJEP3S2/graph.json","fetch_events":"https://pith.science/api/pith-number/J2PB2LJGUNORKN55BTDPJEP3S2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2/action/storage_attestation","attest_author":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2/action/author_attestation","sign_citation":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2/action/citation_signature","submit_replication":"https://pith.science/pith/J2PB2LJGUNORKN55BTDPJEP3S2/action/replication_record"}},"created_at":"2026-07-05T11:10:42.802280+00:00","updated_at":"2026-07-05T11:10:42.802280+00:00"}