{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:MSGLTVDHTDTTGGMBOBEJKUMXPJ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"5d6519813cdefe53c8391ff14f231a7e8f76885fda52c8aade85a3dd58239ded","cross_cats_sorted":["cs.AI","cs.CR","cs.LG"],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-04-08T13:59:02Z","title_canon_sha256":"7b3ec37d0a837b1e99b0ffc0f5b39b99c3c48f98271d4e693047055931ae77ae"},"schema_version":"1.0","source":{"id":"2404.05530","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2404.05530","created_at":"2026-07-05T08:52:31Z"},{"alias_kind":"arxiv_version","alias_value":"2404.05530v2","created_at":"2026-07-05T08:52:31Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.05530","created_at":"2026-07-05T08:52:31Z"},{"alias_kind":"pith_short_12","alias_value":"MSGLTVDHTDTT","created_at":"2026-07-05T08:52:31Z"},{"alias_kind":"pith_short_16","alias_value":"MSGLTVDHTDTTGGMB","created_at":"2026-07-05T08:52:31Z"},{"alias_kind":"pith_short_8","alias_value":"MSGLTVDH","created_at":"2026-07-05T08:52:31Z"}],"graph_snapshots":[{"event_id":"sha256:3c9b2d5a8be00cfedeff78025b8cc5f3af702a7b0325e671367a44de549223b2","target":"graph","created_at":"2026-07-05T08:52:31Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2404.05530/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) is a popular method for aligning Language Models (LM) with human values and preferences. RLHF requires a large number of preference pairs as training data, which are often used in both the Supervised Fine-Tuning and Reward Model training and therefore publicly available datasets are commonly used. In this work, we study to what extent a malicious actor can manipulate the LMs generations by poisoning the preferences, i.e., injecting poisonous preference pairs into these datasets and the RLHF training process. We propose strategies to build poiso","authors_text":"Dana Alon, Donald Metzler, Tim Baumg\\\"artner, Yang Gao","cross_cats":["cs.AI","cs.CR","cs.LG"],"headline":"","license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-04-08T13:59:02Z","title":"Best-of-Venom: Attacking RLHF by Injecting Poisoned Preference Data"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.05530","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d7fd09615d56cec2c433fbe997ab7b9facfee86d23974fa7ef9d34bd43ad5900","target":"record","created_at":"2026-07-05T08:52:31Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"5d6519813cdefe53c8391ff14f231a7e8f76885fda52c8aade85a3dd58239ded","cross_cats_sorted":["cs.AI","cs.CR","cs.LG"],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-04-08T13:59:02Z","title_canon_sha256":"7b3ec37d0a837b1e99b0ffc0f5b39b99c3c48f98271d4e693047055931ae77ae"},"schema_version":"1.0","source":{"id":"2404.05530","kind":"arxiv","version":2}},"canonical_sha256":"648cb9d46798e733198170489551977a6a08c2cccbda2501f271797324e96162","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"648cb9d46798e733198170489551977a6a08c2cccbda2501f271797324e96162","first_computed_at":"2026-07-05T08:52:31.529504Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T08:52:31.529504Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"EnInu7I+VlAcH0nb8tCWZh8lKuRtMwWOLvNozkyND52afPnFiCPGQszIqFD8wAMyiJzEz75/Db92fQj063esAA==","signature_status":"signed_v1","signed_at":"2026-07-05T08:52:31.529907Z","signed_message":"canonical_sha256_bytes"},"source_id":"2404.05530","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d7fd09615d56cec2c433fbe997ab7b9facfee86d23974fa7ef9d34bd43ad5900","sha256:3c9b2d5a8be00cfedeff78025b8cc5f3af702a7b0325e671367a44de549223b2"],"state_sha256":"146826727d7fdcd413d79ec5da59375580dc309b8d6d744cba3fcf39bf3dd6a9"}