{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:WA4HPUYAFG4223QK47MXKIF6RE","short_pith_number":"pith:WA4HPUYA","schema_version":"1.0","canonical_sha256":"b03877d30029b9ad6e0ae7d97520be891bddaf91662d0832f3909e11b356cf50","source":{"kind":"arxiv","id":"2604.15851","version":3},"attestation_state":"computed","paper":{"title":"DPrivBench: Benchmarking LLMs' Reasoning for Differential Privacy","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"DPrivBench shows that top LLMs handle basic differential privacy mechanisms but fail on advanced algorithms, exposing gaps in automated DP reasoning.","cross_cats":["cs.AI","cs.CR"],"primary_cat":"cs.LG","authors_text":"Eli Chien, Erchi Wang, Kamalika Chaudhuri, Om Thakkar, Pengrun Huang, Ruihan Wu, Yu-Xiang Wang","submitted_at":"2026-04-17T09:03:11Z","abstract_excerpt":"Differential privacy (DP) has a wide range of applications for protecting data privacy, but designing and verifying DP algorithms requires expert-level reasoning, creating a high barrier for non-expert practitioners. Prior works either rely on specialized verification languages that demand substantial domain expertise or remain semi-automated and require human-in-the-loop guidance. In this work, we investigate whether large language models (LLMs) can automate DP reasoning. We introduce DPrivBench, a benchmark in which each instance asks whether a function or algorithm satisfies a stated DP gua"},"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":"2604.15851","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-04-17T09:03:11Z","cross_cats_sorted":["cs.AI","cs.CR"],"title_canon_sha256":"7c00df388e9485eac7a07815c9502594e2c8e4d5b1d3171f8a9a695f793a4c53","abstract_canon_sha256":"f99ea8b91d69c7290d1d304f8b51b02d60a37b0c55433ed60a44d39faf7ffc6f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:31.798745Z","signature_b64":"D9ojmYg7meMJGS91ObxhcmBjOZ4EfKgv0gjAVDM/syK+6CK3nCqnNygEwRnJOHZaOcPgWkWX6Jaugq8x3souBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b03877d30029b9ad6e0ae7d97520be891bddaf91662d0832f3909e11b356cf50","last_reissued_at":"2026-05-20T00:04:31.797946Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:31.797946Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DPrivBench: Benchmarking LLMs' Reasoning for Differential Privacy","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"DPrivBench shows that top LLMs handle basic differential privacy mechanisms but fail on advanced algorithms, exposing gaps in automated DP reasoning.","cross_cats":["cs.AI","cs.CR"],"primary_cat":"cs.LG","authors_text":"Eli Chien, Erchi Wang, Kamalika Chaudhuri, Om Thakkar, Pengrun Huang, Ruihan Wu, Yu-Xiang Wang","submitted_at":"2026-04-17T09:03:11Z","abstract_excerpt":"Differential privacy (DP) has a wide range of applications for protecting data privacy, but designing and verifying DP algorithms requires expert-level reasoning, creating a high barrier for non-expert practitioners. Prior works either rely on specialized verification languages that demand substantial domain expertise or remain semi-automated and require human-in-the-loop guidance. In this work, we investigate whether large language models (LLMs) can automate DP reasoning. We introduce DPrivBench, a benchmark in which each instance asks whether a function or algorithm satisfies a stated DP gua"},"claims":{"count":3,"items":[{"kind":"strongest_claim","text":"Experiments show that while the strongest models handle textbook mechanisms well, all models struggle with advanced algorithms, revealing substantial gaps in current DP reasoning capabilities.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The benchmark instances are free of shortcut reasoning patterns and correctly labeled as to whether each function satisfies the stated DP guarantee under the given assumptions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DPrivBench shows that top LLMs handle basic differential privacy mechanisms but fail on advanced algorithms, exposing gaps in automated DP reasoning.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"}],"snapshot_sha256":"bc06401d08860c7ce4046d956490a23b1fb65525ac3995b77a1e288b72dcbe24"},"source":{"id":"2604.15851","kind":"arxiv","version":3},"verdict":{"id":"12abc099-38f1-4d72-9bc4-c7ee43de3d34","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T08:50:47.718349Z","strongest_claim":"Experiments show that while the strongest models handle textbook mechanisms well, all models struggle with advanced algorithms, revealing substantial gaps in current DP reasoning capabilities.","one_line_summary":"DPrivBench shows that top LLMs handle basic differential privacy mechanisms but fail on advanced algorithms, exposing gaps in automated DP reasoning.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The benchmark instances are free of shortcut reasoning patterns and correctly labeled as to whether each function satisfies the stated DP guarantee under the given assumptions.","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.15851/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":"2604.15851","created_at":"2026-05-20T00:04:31.798078+00:00"},{"alias_kind":"arxiv_version","alias_value":"2604.15851v3","created_at":"2026-05-20T00:04:31.798078+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.15851","created_at":"2026-05-20T00:04:31.798078+00:00"},{"alias_kind":"pith_short_12","alias_value":"WA4HPUYAFG42","created_at":"2026-05-20T00:04:31.798078+00:00"},{"alias_kind":"pith_short_16","alias_value":"WA4HPUYAFG4223QK","created_at":"2026-05-20T00:04:31.798078+00:00"},{"alias_kind":"pith_short_8","alias_value":"WA4HPUYA","created_at":"2026-05-20T00:04:31.798078+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/WA4HPUYAFG4223QK47MXKIF6RE","json":"https://pith.science/pith/WA4HPUYAFG4223QK47MXKIF6RE.json","graph_json":"https://pith.science/api/pith-number/WA4HPUYAFG4223QK47MXKIF6RE/graph.json","events_json":"https://pith.science/api/pith-number/WA4HPUYAFG4223QK47MXKIF6RE/events.json","paper":"https://pith.science/paper/WA4HPUYA"},"agent_actions":{"view_html":"https://pith.science/pith/WA4HPUYAFG4223QK47MXKIF6RE","download_json":"https://pith.science/pith/WA4HPUYAFG4223QK47MXKIF6RE.json","view_paper":"https://pith.science/paper/WA4HPUYA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2604.15851&json=true","fetch_graph":"https://pith.science/api/pith-number/WA4HPUYAFG4223QK47MXKIF6RE/graph.json","fetch_events":"https://pith.science/api/pith-number/WA4HPUYAFG4223QK47MXKIF6RE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WA4HPUYAFG4223QK47MXKIF6RE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WA4HPUYAFG4223QK47MXKIF6RE/action/storage_attestation","attest_author":"https://pith.science/pith/WA4HPUYAFG4223QK47MXKIF6RE/action/author_attestation","sign_citation":"https://pith.science/pith/WA4HPUYAFG4223QK47MXKIF6RE/action/citation_signature","submit_replication":"https://pith.science/pith/WA4HPUYAFG4223QK47MXKIF6RE/action/replication_record"}},"created_at":"2026-05-20T00:04:31.798078+00:00","updated_at":"2026-05-20T00:04:31.798078+00:00"}