{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7WH4LKBMPEWSBDUJQHADZM2RB6","short_pith_number":"pith:7WH4LKBM","schema_version":"1.0","canonical_sha256":"fd8fc5a82c792d208e8981c03cb3510f8fd1a43def4b81f929f5b07741544927","source":{"kind":"arxiv","id":"2608.05040","version":1},"attestation_state":"computed","paper":{"title":"Private Direct Preference Optimization for LLM Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Bolin Ding, Ergute Bao, Fei Wei, Xiaokui Xiao, Yaliang Li, YangFan Jiang","submitted_at":"2026-08-05T16:51:02Z","abstract_excerpt":"Direct preference optimization (DPO) is now a standard method for aligning large language models (LLMs) using human preference data. Each DPO example contains a prompt and a pair of candidate model responses. While prompts and responses are often public or model-generated, the relative preference between responses reflects subjective judgments and can reveal sensitive attributes of annotators or end users. Off-the-shelf privacy-preserving approaches are not well matched to this structure, leading to unnecessary noise injection and biased updates in training.\n  In this paper, we formalize prefe"},"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":"2608.05040","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2026-08-05T16:51:02Z","cross_cats_sorted":[],"title_canon_sha256":"7114c2e0f6fa89893bfd5254d8fbc108c7638d054999a566c8ca611ea81f92cb","abstract_canon_sha256":"f6a17ffe4d1b5ec2a1fdafa7fef4a4c76b74cce48f8a15f09da71e473b9488ff"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-06T01:48:05.705905Z","signature_b64":"O6KItOnAEV7cJUUrSZdR+k/3KGpEmJ3ziEOxlvLiq5zY84ZWxkU95p/0SVBQlm4tUue/NVc8Ix/quA1Y0t4FAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fd8fc5a82c792d208e8981c03cb3510f8fd1a43def4b81f929f5b07741544927","last_reissued_at":"2026-08-06T01:48:05.704483Z","signature_status":"signed_v1","first_computed_at":"2026-08-06T01:48:05.704483Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Private Direct Preference Optimization for LLM Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CR","authors_text":"Bolin Ding, Ergute Bao, Fei Wei, Xiaokui Xiao, Yaliang Li, YangFan Jiang","submitted_at":"2026-08-05T16:51:02Z","abstract_excerpt":"Direct preference optimization (DPO) is now a standard method for aligning large language models (LLMs) using human preference data. Each DPO example contains a prompt and a pair of candidate model responses. While prompts and responses are often public or model-generated, the relative preference between responses reflects subjective judgments and can reveal sensitive attributes of annotators or end users. Off-the-shelf privacy-preserving approaches are not well matched to this structure, leading to unnecessary noise injection and biased updates in training.\n  In this paper, we formalize prefe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.05040","kind":"arxiv","version":1},"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/2608.05040/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":"2608.05040","created_at":"2026-08-06T01:48:05.706338+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.05040v1","created_at":"2026-08-06T01:48:05.706338+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.05040","created_at":"2026-08-06T01:48:05.706338+00:00"},{"alias_kind":"pith_short_12","alias_value":"7WH4LKBMPEWS","created_at":"2026-08-06T01:48:05.706338+00:00"},{"alias_kind":"pith_short_16","alias_value":"7WH4LKBMPEWSBDUJ","created_at":"2026-08-06T01:48:05.706338+00:00"},{"alias_kind":"pith_short_8","alias_value":"7WH4LKBM","created_at":"2026-08-06T01:48:05.706338+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6","json":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6.json","graph_json":"https://pith.science/api/pith-number/7WH4LKBMPEWSBDUJQHADZM2RB6/graph.json","events_json":"https://pith.science/api/pith-number/7WH4LKBMPEWSBDUJQHADZM2RB6/events.json","paper":"https://pith.science/paper/7WH4LKBM"},"agent_actions":{"view_html":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6","download_json":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6.json","view_paper":"https://pith.science/paper/7WH4LKBM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.05040&json=true","fetch_graph":"https://pith.science/api/pith-number/7WH4LKBMPEWSBDUJQHADZM2RB6/graph.json","fetch_events":"https://pith.science/api/pith-number/7WH4LKBMPEWSBDUJQHADZM2RB6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6/action/storage_attestation","attest_author":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6/action/author_attestation","sign_citation":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6/action/citation_signature","submit_replication":"https://pith.science/pith/7WH4LKBMPEWSBDUJQHADZM2RB6/action/replication_record"}},"created_at":"2026-08-06T01:48:05.706338+00:00","updated_at":"2026-08-06T01:48:05.706338+00:00"}