{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:OJHFYRGOQDFYR5SN7GF5GLL4KE","short_pith_number":"pith:OJHFYRGO","schema_version":"1.0","canonical_sha256":"724e5c44ce80cb88f64df98bd32d7c5101aa844586b927cf85519f842508b0ca","source":{"kind":"arxiv","id":"2402.10892","version":3},"attestation_state":"computed","paper":{"title":"Proving membership in LLM pretraining data via data watermarks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Johnny Tian-Zheng Wei, Robin Jia, Ryan Yixiang Wang","submitted_at":"2024-02-16T18:49:27Z","abstract_excerpt":"Detecting whether copyright holders' works were used in LLM pretraining is poised to be an important problem. This work proposes using data watermarks to enable principled detection with only black-box model access, provided that the rightholder contributed multiple training documents and watermarked them before public release. By applying a randomly sampled data watermark, detection can be framed as hypothesis testing, which provides guarantees on the false detection rate. We study two watermarks: one that inserts random sequences, and another that randomly substitutes characters with Unicode"},"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":"2402.10892","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2024-02-16T18:49:27Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"e94a2a340628604fba43514747157e1e192a73f24722027eabcdd57b3fdef6a4","abstract_canon_sha256":"ed43342ded91d2e0c53fbc64be11c4fceb79d3998714cff2ea0354ce98982778"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:56:17.222965Z","signature_b64":"TFdGRlHYGcaX7/UPGdbdkNvws7VZTR2uD5yWdhgWLz1YaHILQs/EtYtvHh9nfB5l1GePTS7a1p85DK3s0OsgBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"724e5c44ce80cb88f64df98bd32d7c5101aa844586b927cf85519f842508b0ca","last_reissued_at":"2026-07-05T08:56:17.222487Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:56:17.222487Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Proving membership in LLM pretraining data via data watermarks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Johnny Tian-Zheng Wei, Robin Jia, Ryan Yixiang Wang","submitted_at":"2024-02-16T18:49:27Z","abstract_excerpt":"Detecting whether copyright holders' works were used in LLM pretraining is poised to be an important problem. This work proposes using data watermarks to enable principled detection with only black-box model access, provided that the rightholder contributed multiple training documents and watermarked them before public release. By applying a randomly sampled data watermark, detection can be framed as hypothesis testing, which provides guarantees on the false detection rate. We study two watermarks: one that inserts random sequences, and another that randomly substitutes characters with Unicode"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.10892","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/2402.10892/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":"2402.10892","created_at":"2026-07-05T08:56:17.222545+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.10892v3","created_at":"2026-07-05T08:56:17.222545+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.10892","created_at":"2026-07-05T08:56:17.222545+00:00"},{"alias_kind":"pith_short_12","alias_value":"OJHFYRGOQDFY","created_at":"2026-07-05T08:56:17.222545+00:00"},{"alias_kind":"pith_short_16","alias_value":"OJHFYRGOQDFYR5SN","created_at":"2026-07-05T08:56:17.222545+00:00"},{"alias_kind":"pith_short_8","alias_value":"OJHFYRGO","created_at":"2026-07-05T08:56:17.222545+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2510.18333","citing_title":"Position: LLM Watermarking Should Align Stakeholders' Incentives for Practical Adoption","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03547","citing_title":"Erase Persona, Forget Lore: Benchmarking Multimodal Copyright Unlearning in Large Vision Language Models","ref_index":31,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE","json":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE.json","graph_json":"https://pith.science/api/pith-number/OJHFYRGOQDFYR5SN7GF5GLL4KE/graph.json","events_json":"https://pith.science/api/pith-number/OJHFYRGOQDFYR5SN7GF5GLL4KE/events.json","paper":"https://pith.science/paper/OJHFYRGO"},"agent_actions":{"view_html":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE","download_json":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE.json","view_paper":"https://pith.science/paper/OJHFYRGO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.10892&json=true","fetch_graph":"https://pith.science/api/pith-number/OJHFYRGOQDFYR5SN7GF5GLL4KE/graph.json","fetch_events":"https://pith.science/api/pith-number/OJHFYRGOQDFYR5SN7GF5GLL4KE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE/action/storage_attestation","attest_author":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE/action/author_attestation","sign_citation":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE/action/citation_signature","submit_replication":"https://pith.science/pith/OJHFYRGOQDFYR5SN7GF5GLL4KE/action/replication_record"}},"created_at":"2026-07-05T08:56:17.222545+00:00","updated_at":"2026-07-05T08:56:17.222545+00:00"}