{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:IWTD6VQBPBVA35HAABYX6WXY4N","short_pith_number":"pith:IWTD6VQB","schema_version":"1.0","canonical_sha256":"45a63f5601786a0df4e000717f5af8e372157ab786d19e66b765035e568e1032","source":{"kind":"arxiv","id":"2502.18290","version":3},"attestation_state":"computed","paper":{"title":"Stealthy Backdoor Attack in Self-Supervised Learning Vision Encoders for Large Vision Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Huan Zhang, Zhaoyi Liu","submitted_at":"2025-02-25T15:28:41Z","abstract_excerpt":"Self-supervised learning (SSL) vision encoders learn high-quality image representations and thus have become a vital part of developing vision modality of large vision language models (LVLMs). Due to the high cost of training such encoders, pre-trained encoders are widely shared and deployed into many LVLMs, which are security-critical or bear societal significance. Under this practical scenario, we reveal a new backdoor threat that significant visual hallucinations can be induced into these LVLMs by merely compromising vision encoders. Because of the sharing and reuse of these encoders, many "},"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":"2502.18290","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-02-25T15:28:41Z","cross_cats_sorted":[],"title_canon_sha256":"fa83f9299175a9df4cf48805e5cb4e4d0827736d9b6005496363d98222764f00","abstract_canon_sha256":"7f2659886ae54f0a0fd14d1a4e6f91204daac9cea34fa65899e80d5c55caf593"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:39:26.421219Z","signature_b64":"E+kjtw1nslgsIWbXMc3bC2oceyNtGo0ZkOCKbXE3KONzdaz8sSZHX9UcHzWe3zszqtGffqjl8kcta+oZ+eVSAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"45a63f5601786a0df4e000717f5af8e372157ab786d19e66b765035e568e1032","last_reissued_at":"2026-07-05T10:39:26.420773Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:39:26.420773Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stealthy Backdoor Attack in Self-Supervised Learning Vision Encoders for Large Vision Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Huan Zhang, Zhaoyi Liu","submitted_at":"2025-02-25T15:28:41Z","abstract_excerpt":"Self-supervised learning (SSL) vision encoders learn high-quality image representations and thus have become a vital part of developing vision modality of large vision language models (LVLMs). Due to the high cost of training such encoders, pre-trained encoders are widely shared and deployed into many LVLMs, which are security-critical or bear societal significance. Under this practical scenario, we reveal a new backdoor threat that significant visual hallucinations can be induced into these LVLMs by merely compromising vision encoders. Because of the sharing and reuse of these encoders, many "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.18290","kind":"arxiv","version":3},"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/2502.18290/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":"2502.18290","created_at":"2026-07-05T10:39:26.420825+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.18290v3","created_at":"2026-07-05T10:39:26.420825+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.18290","created_at":"2026-07-05T10:39:26.420825+00:00"},{"alias_kind":"pith_short_12","alias_value":"IWTD6VQBPBVA","created_at":"2026-07-05T10:39:26.420825+00:00"},{"alias_kind":"pith_short_16","alias_value":"IWTD6VQBPBVA35HA","created_at":"2026-07-05T10:39:26.420825+00:00"},{"alias_kind":"pith_short_8","alias_value":"IWTD6VQB","created_at":"2026-07-05T10:39:26.420825+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.07214","citing_title":"Backdoor Attack on Vision Language Models with Stealthy Semantic Manipulation","ref_index":5,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N","json":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N.json","graph_json":"https://pith.science/api/pith-number/IWTD6VQBPBVA35HAABYX6WXY4N/graph.json","events_json":"https://pith.science/api/pith-number/IWTD6VQBPBVA35HAABYX6WXY4N/events.json","paper":"https://pith.science/paper/IWTD6VQB"},"agent_actions":{"view_html":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N","download_json":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N.json","view_paper":"https://pith.science/paper/IWTD6VQB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.18290&json=true","fetch_graph":"https://pith.science/api/pith-number/IWTD6VQBPBVA35HAABYX6WXY4N/graph.json","fetch_events":"https://pith.science/api/pith-number/IWTD6VQBPBVA35HAABYX6WXY4N/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N/action/storage_attestation","attest_author":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N/action/author_attestation","sign_citation":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N/action/citation_signature","submit_replication":"https://pith.science/pith/IWTD6VQBPBVA35HAABYX6WXY4N/action/replication_record"}},"created_at":"2026-07-05T10:39:26.420825+00:00","updated_at":"2026-07-05T10:39:26.420825+00:00"}