{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:EDVLW45EP2OGOPBIZ3HXB3O7XT","short_pith_number":"pith:EDVLW45E","schema_version":"1.0","canonical_sha256":"20eabb73a47e9c673c28cecf70eddfbcf89ba79995fd005c730e3d25dd4d3380","source":{"kind":"arxiv","id":"2302.10149","version":2},"attestation_state":"computed","paper":{"title":"Poisoning Web-Scale Training Datasets is Practical","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Andreas Terzis, Christopher A. Choquette-Choo, Daniel Paleka, Florian Tram\\`er, Hyrum Anderson, Kurt Thomas, Matthew Jagielski, Nicholas Carlini, Will Pearce","submitted_at":"2023-02-20T18:30:54Z","abstract_excerpt":"Deep learning models are often trained on distributed, web-scale datasets crawled from the internet. In this paper, we introduce two new dataset poisoning attacks that intentionally introduce malicious examples to a model's performance. Our attacks are immediately practical and could, today, poison 10 popular datasets. Our first attack, split-view poisoning, exploits the mutable nature of internet content to ensure a dataset annotator's initial view of the dataset differs from the view downloaded by subsequent clients. By exploiting specific invalid trust assumptions, we show how we could have"},"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":"2302.10149","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2023-02-20T18:30:54Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"0b8ccb44a52cab3c6f5f3562c39da2e205d30b57c3997cc262b356c271763755","abstract_canon_sha256":"4dc41654801322c16873eb96e51e5e5ea4a9071dd043d9bfa5f2347aa237353d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:15:35.087600Z","signature_b64":"rJM5eKo99bTFcQaYeGcVDWx2Z6VlL46rkklr9EuSRMQLAp+tv4mBOwMRQni5ocVERemxCWqzEXoY6sM8OX8KCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"20eabb73a47e9c673c28cecf70eddfbcf89ba79995fd005c730e3d25dd4d3380","last_reissued_at":"2026-07-05T08:15:35.086663Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:15:35.086663Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Poisoning Web-Scale Training Datasets is Practical","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Andreas Terzis, Christopher A. Choquette-Choo, Daniel Paleka, Florian Tram\\`er, Hyrum Anderson, Kurt Thomas, Matthew Jagielski, Nicholas Carlini, Will Pearce","submitted_at":"2023-02-20T18:30:54Z","abstract_excerpt":"Deep learning models are often trained on distributed, web-scale datasets crawled from the internet. In this paper, we introduce two new dataset poisoning attacks that intentionally introduce malicious examples to a model's performance. Our attacks are immediately practical and could, today, poison 10 popular datasets. Our first attack, split-view poisoning, exploits the mutable nature of internet content to ensure a dataset annotator's initial view of the dataset differs from the view downloaded by subsequent clients. By exploiting specific invalid trust assumptions, we show how we could have"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2302.10149","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/2302.10149/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":"2302.10149","created_at":"2026-07-05T08:15:35.086733+00:00"},{"alias_kind":"arxiv_version","alias_value":"2302.10149v2","created_at":"2026-07-05T08:15:35.086733+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2302.10149","created_at":"2026-07-05T08:15:35.086733+00:00"},{"alias_kind":"pith_short_12","alias_value":"EDVLW45EP2OG","created_at":"2026-07-05T08:15:35.086733+00:00"},{"alias_kind":"pith_short_16","alias_value":"EDVLW45EP2OGOPBI","created_at":"2026-07-05T08:15:35.086733+00:00"},{"alias_kind":"pith_short_8","alias_value":"EDVLW45E","created_at":"2026-07-05T08:15:35.086733+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01280","citing_title":"Fixed-Set Robustness in Programming by Example: Example Corruption and Semantic Partition Recovery","ref_index":62,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00279","citing_title":"Bit-Exact AI Inference Verification Without Performance Tradeoffs","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2606.32002","citing_title":"Self-Study Reconsidered: The Hidden Fragility of Learning from Self-Generated QA","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2606.24890","citing_title":"Small edits, large models: How Wikipedia advocacy shapes LLM values","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06846","citing_title":"Narrow Secret Loyalty Dodges Black-Box Audits","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2404.02696","citing_title":"Deep Privacy Funnel Model: From a Discriminative to a Generative Approach with an Application to Face Recognition","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21860","citing_title":"Robust Statistical Estimators with Bounded Empirical Sensitivity","ref_index":283,"is_internal_anchor":false},{"citing_arxiv_id":"2305.17493","citing_title":"The Curse of Recursion: Training on Generated Data Makes Models Forget","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2406.10162","citing_title":"Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06846","citing_title":"Narrow Secret Loyalty Dodges Black-Box Audits","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06846","citing_title":"Narrow Secret Loyalty Dodges Black-Box Audits","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT","json":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT.json","graph_json":"https://pith.science/api/pith-number/EDVLW45EP2OGOPBIZ3HXB3O7XT/graph.json","events_json":"https://pith.science/api/pith-number/EDVLW45EP2OGOPBIZ3HXB3O7XT/events.json","paper":"https://pith.science/paper/EDVLW45E"},"agent_actions":{"view_html":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT","download_json":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT.json","view_paper":"https://pith.science/paper/EDVLW45E","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2302.10149&json=true","fetch_graph":"https://pith.science/api/pith-number/EDVLW45EP2OGOPBIZ3HXB3O7XT/graph.json","fetch_events":"https://pith.science/api/pith-number/EDVLW45EP2OGOPBIZ3HXB3O7XT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT/action/storage_attestation","attest_author":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT/action/author_attestation","sign_citation":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT/action/citation_signature","submit_replication":"https://pith.science/pith/EDVLW45EP2OGOPBIZ3HXB3O7XT/action/replication_record"}},"created_at":"2026-07-05T08:15:35.086733+00:00","updated_at":"2026-07-05T08:15:35.086733+00:00"}