{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:6434FX4X2M7YED6H5I5EALO5HD","short_pith_number":"pith:6434FX4X","schema_version":"1.0","canonical_sha256":"f737c2df97d33f820fc7ea3a402ddd38fe0228a3dd4bfff867bfc1301062533d","source":{"kind":"arxiv","id":"2010.12563","version":2},"attestation_state":"computed","paper":{"title":"Concealed Data Poisoning Attacks on NLP Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Eric Wallace, Sameer Singh, Shi Feng, Tony Z. Zhao","submitted_at":"2020-10-23T17:47:06Z","abstract_excerpt":"Adversarial attacks alter NLP model predictions by perturbing test-time inputs. However, it is much less understood whether, and how, predictions can be manipulated with small, concealed changes to the training data. In this work, we develop a new data poisoning attack that allows an adversary to control model predictions whenever a desired trigger phrase is present in the input. For instance, we insert 50 poison examples into a sentiment model's training set that causes the model to frequently predict Positive whenever the input contains \"James Bond\". Crucially, we craft these poison examples"},"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":"2010.12563","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-10-23T17:47:06Z","cross_cats_sorted":[],"title_canon_sha256":"bd774612ee1c42201cc63cdd259cae22e60e61e36797af46bbe3b81d5d4c6c6b","abstract_canon_sha256":"a6d00316afe7097588692d7ea01700c37762761b3d97f16e35c0b19f7101f847"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:31:00.531062Z","signature_b64":"Rq1t2OylbGAT6rDP4Vp19NN3l9DSeZtKDOolzYvbHNAiv/CrL4i1XfbPZbdhzaz7Xj4o3AKievmiZO4Cnyp2Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f737c2df97d33f820fc7ea3a402ddd38fe0228a3dd4bfff867bfc1301062533d","last_reissued_at":"2026-07-05T02:31:00.530543Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:31:00.530543Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Concealed Data Poisoning Attacks on NLP Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Eric Wallace, Sameer Singh, Shi Feng, Tony Z. Zhao","submitted_at":"2020-10-23T17:47:06Z","abstract_excerpt":"Adversarial attacks alter NLP model predictions by perturbing test-time inputs. However, it is much less understood whether, and how, predictions can be manipulated with small, concealed changes to the training data. In this work, we develop a new data poisoning attack that allows an adversary to control model predictions whenever a desired trigger phrase is present in the input. For instance, we insert 50 poison examples into a sentiment model's training set that causes the model to frequently predict Positive whenever the input contains \"James Bond\". Crucially, we craft these poison examples"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2010.12563","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/2010.12563/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":"2010.12563","created_at":"2026-07-05T02:31:00.530602+00:00"},{"alias_kind":"arxiv_version","alias_value":"2010.12563v2","created_at":"2026-07-05T02:31:00.530602+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2010.12563","created_at":"2026-07-05T02:31:00.530602+00:00"},{"alias_kind":"pith_short_12","alias_value":"6434FX4X2M7Y","created_at":"2026-07-05T02:31:00.530602+00:00"},{"alias_kind":"pith_short_16","alias_value":"6434FX4X2M7YED6H","created_at":"2026-07-05T02:31:00.530602+00:00"},{"alias_kind":"pith_short_8","alias_value":"6434FX4X","created_at":"2026-07-05T02:31:00.530602+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25476","citing_title":"A Red Teaming Framework for Large Language Models: A Case Study on Faithfulness Evaluation","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2512.04457","citing_title":"RapidUn: Influence-Driven Parameter Reweighting for Efficient Large Language Model Unlearning","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD","json":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD.json","graph_json":"https://pith.science/api/pith-number/6434FX4X2M7YED6H5I5EALO5HD/graph.json","events_json":"https://pith.science/api/pith-number/6434FX4X2M7YED6H5I5EALO5HD/events.json","paper":"https://pith.science/paper/6434FX4X"},"agent_actions":{"view_html":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD","download_json":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD.json","view_paper":"https://pith.science/paper/6434FX4X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2010.12563&json=true","fetch_graph":"https://pith.science/api/pith-number/6434FX4X2M7YED6H5I5EALO5HD/graph.json","fetch_events":"https://pith.science/api/pith-number/6434FX4X2M7YED6H5I5EALO5HD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD/action/storage_attestation","attest_author":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD/action/author_attestation","sign_citation":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD/action/citation_signature","submit_replication":"https://pith.science/pith/6434FX4X2M7YED6H5I5EALO5HD/action/replication_record"}},"created_at":"2026-07-05T02:31:00.530602+00:00","updated_at":"2026-07-05T02:31:00.530602+00:00"}