{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:77SGTNLDDQYZ5KK7JS2XHG72DT","short_pith_number":"pith:77SGTNLD","schema_version":"1.0","canonical_sha256":"ffe469b5631c319ea95f4cb5739bfa1cc1a3abc8f11f2a713e3c5e837f700caa","source":{"kind":"arxiv","id":"2408.02946","version":6},"attestation_state":"computed","paper":{"title":"Scaling Trends for Data Poisoning in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CR","authors_text":"Adam Gleave, Brendan Murphy, David Khachaturov, Dillon Bowen, Kellin Pelrine, Will Cai","submitted_at":"2024-08-06T04:14:29Z","abstract_excerpt":"LLMs produce harmful and undesirable behavior when trained on datasets containing even a small fraction of poisoned data. We demonstrate that GPT models remain vulnerable to fine-tuning on poisoned data, even when safeguarded by moderation systems. Given the persistence of data poisoning vulnerabilities in today's most capable models, this paper investigates whether these risks increase with model scaling. We evaluate three threat models -- malicious fine-tuning, imperfect data curation, and intentional data contamination -- across 24 frontier LLMs ranging from 1.5 to 72 billion parameters. Ou"},"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":"2408.02946","kind":"arxiv","version":6},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2024-08-06T04:14:29Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"9ca22be05ad4cc9c0f9bef15738e75771feded96c2cdbcab4164198e80a3328f","abstract_canon_sha256":"81f1a7f9000c956d37406f828d0ea31b763f901b9ac6b68fe6798222f605c63e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:38:23.799912Z","signature_b64":"xBRpTYzen1U+rO1pTQobSkMrGHhPwqXJVL+WEfxQh8ARjk4aekTojn1kHpBGPLMLG1pnnNptmFg7wr5mvuuiDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ffe469b5631c319ea95f4cb5739bfa1cc1a3abc8f11f2a713e3c5e837f700caa","last_reissued_at":"2026-07-05T11:38:23.799310Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:38:23.799310Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Scaling Trends for Data Poisoning in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CR","authors_text":"Adam Gleave, Brendan Murphy, David Khachaturov, Dillon Bowen, Kellin Pelrine, Will Cai","submitted_at":"2024-08-06T04:14:29Z","abstract_excerpt":"LLMs produce harmful and undesirable behavior when trained on datasets containing even a small fraction of poisoned data. We demonstrate that GPT models remain vulnerable to fine-tuning on poisoned data, even when safeguarded by moderation systems. Given the persistence of data poisoning vulnerabilities in today's most capable models, this paper investigates whether these risks increase with model scaling. We evaluate three threat models -- malicious fine-tuning, imperfect data curation, and intentional data contamination -- across 24 frontier LLMs ranging from 1.5 to 72 billion parameters. Ou"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.02946","kind":"arxiv","version":6},"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/2408.02946/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":"2408.02946","created_at":"2026-07-05T11:38:23.799422+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.02946v6","created_at":"2026-07-05T11:38:23.799422+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.02946","created_at":"2026-07-05T11:38:23.799422+00:00"},{"alias_kind":"pith_short_12","alias_value":"77SGTNLDDQYZ","created_at":"2026-07-05T11:38:23.799422+00:00"},{"alias_kind":"pith_short_16","alias_value":"77SGTNLDDQYZ5KK7","created_at":"2026-07-05T11:38:23.799422+00:00"},{"alias_kind":"pith_short_8","alias_value":"77SGTNLD","created_at":"2026-07-05T11:38:23.799422+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.00036","citing_title":"AI Integrity: Defending Against Backdoors and Secret Loyalties","ref_index":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT","json":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT.json","graph_json":"https://pith.science/api/pith-number/77SGTNLDDQYZ5KK7JS2XHG72DT/graph.json","events_json":"https://pith.science/api/pith-number/77SGTNLDDQYZ5KK7JS2XHG72DT/events.json","paper":"https://pith.science/paper/77SGTNLD"},"agent_actions":{"view_html":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT","download_json":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT.json","view_paper":"https://pith.science/paper/77SGTNLD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.02946&json=true","fetch_graph":"https://pith.science/api/pith-number/77SGTNLDDQYZ5KK7JS2XHG72DT/graph.json","fetch_events":"https://pith.science/api/pith-number/77SGTNLDDQYZ5KK7JS2XHG72DT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT/action/storage_attestation","attest_author":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT/action/author_attestation","sign_citation":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT/action/citation_signature","submit_replication":"https://pith.science/pith/77SGTNLDDQYZ5KK7JS2XHG72DT/action/replication_record"}},"created_at":"2026-07-05T11:38:23.799422+00:00","updated_at":"2026-07-05T11:38:23.799422+00:00"}