{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7PIIWTD3NE4IIGR62R7VWGV4HZ","short_pith_number":"pith:7PIIWTD3","schema_version":"1.0","canonical_sha256":"fbd08b4c7b6938841a3ed47f5b1abc3e5147586d82da66d8f3b7153f622ea357","source":{"kind":"arxiv","id":"2606.10904","version":1},"attestation_state":"computed","paper":{"title":"Comparative Analysis of Inference-Time Defense Methods for Multimodal Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Bulat Nutfullin, Dmitry Namiot, Vladimir Evgrafov","submitted_at":"2026-06-09T14:13:54Z","abstract_excerpt":"Multimodal large language models (MLLMs) now appear in safety-critical applications, but the visual channel leaves them open to adversarial attacks that predominantly text-oriented safety alignment addresses only in part. Retraining a model for each new vulnerability class is usually too expensive to be practical. We report a comparative empirical evaluation of three inference-time defense methods and their combinations, run on eight models from the InternVL and Qwen-VL families across seven safety benchmarks that span four attack classes and total 9,000 evaluation samples. Every figure below "},"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":"2606.10904","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2026-06-09T14:13:54Z","cross_cats_sorted":[],"title_canon_sha256":"a880486fbdfddd4e76a67e0262d98063ce792aa104d7e48a9e990e38ae910824","abstract_canon_sha256":"1012409313d5dbcef3c2de2e0ff117f772e2828342ce60ee6dcbd3990e919b21"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-10T01:10:47.260480Z","signature_b64":"BGSQpgCDm9c1ck8k3jySKRzetaavhfeMqU1LW5S8DtQWDrANFc3Tz7ykW8Wd3/r0hXq1RdBfwPSprWpYN2lwDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fbd08b4c7b6938841a3ed47f5b1abc3e5147586d82da66d8f3b7153f622ea357","last_reissued_at":"2026-06-10T01:10:47.259683Z","signature_status":"signed_v1","first_computed_at":"2026-06-10T01:10:47.259683Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Comparative Analysis of Inference-Time Defense Methods for Multimodal Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Bulat Nutfullin, Dmitry Namiot, Vladimir Evgrafov","submitted_at":"2026-06-09T14:13:54Z","abstract_excerpt":"Multimodal large language models (MLLMs) now appear in safety-critical applications, but the visual channel leaves them open to adversarial attacks that predominantly text-oriented safety alignment addresses only in part. Retraining a model for each new vulnerability class is usually too expensive to be practical. We report a comparative empirical evaluation of three inference-time defense methods and their combinations, run on eight models from the InternVL and Qwen-VL families across seven safety benchmarks that span four attack classes and total 9,000 evaluation samples. Every figure below "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.10904","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/2606.10904/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":"2606.10904","created_at":"2026-06-10T01:10:47.259823+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.10904v1","created_at":"2026-06-10T01:10:47.259823+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.10904","created_at":"2026-06-10T01:10:47.259823+00:00"},{"alias_kind":"pith_short_12","alias_value":"7PIIWTD3NE4I","created_at":"2026-06-10T01:10:47.259823+00:00"},{"alias_kind":"pith_short_16","alias_value":"7PIIWTD3NE4IIGR6","created_at":"2026-06-10T01:10:47.259823+00:00"},{"alias_kind":"pith_short_8","alias_value":"7PIIWTD3","created_at":"2026-06-10T01:10:47.259823+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ","json":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ.json","graph_json":"https://pith.science/api/pith-number/7PIIWTD3NE4IIGR62R7VWGV4HZ/graph.json","events_json":"https://pith.science/api/pith-number/7PIIWTD3NE4IIGR62R7VWGV4HZ/events.json","paper":"https://pith.science/paper/7PIIWTD3"},"agent_actions":{"view_html":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ","download_json":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ.json","view_paper":"https://pith.science/paper/7PIIWTD3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.10904&json=true","fetch_graph":"https://pith.science/api/pith-number/7PIIWTD3NE4IIGR62R7VWGV4HZ/graph.json","fetch_events":"https://pith.science/api/pith-number/7PIIWTD3NE4IIGR62R7VWGV4HZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ/action/storage_attestation","attest_author":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ/action/author_attestation","sign_citation":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ/action/citation_signature","submit_replication":"https://pith.science/pith/7PIIWTD3NE4IIGR62R7VWGV4HZ/action/replication_record"}},"created_at":"2026-06-10T01:10:47.259823+00:00","updated_at":"2026-06-10T01:10:47.259823+00:00"}