{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VIGEA2AM4XTL43FDYYCJFFBNW2","short_pith_number":"pith:VIGEA2AM","schema_version":"1.0","canonical_sha256":"aa0c40680ce5e6be6ca3c60492942db6971197ca337366b5312d17526301222e","source":{"kind":"arxiv","id":"2401.14446","version":3},"attestation_state":"computed","paper":{"title":"Black-Box Access is Insufficient for Rigorous AI Audits","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CR"],"primary_cat":"cs.CY","authors_text":"Alan Chan, Andreas Haupt, Benjamin Bucknall, Carson Ezell, Charlotte Siegmann, David Bau, David Krueger, Dylan Hadfield-Menell, J\\'er\\'emy Scheurer, Kevin Wei, Lee Sharkey, Marius Hobbhahn, Marvin Von Hagen, Max Tegmark, Michael Gerovitch, Noam Kolt, Qinyi Sun, Satyapriya Krishna, Silas Alberti, Stephen Casper, Taylor Lynn Curtis","submitted_at":"2024-01-25T18:58:05Z","abstract_excerpt":"External audits of AI systems are increasingly recognized as a key mechanism for AI governance. The effectiveness of an audit, however, depends on the degree of access granted to auditors. Recent audits of state-of-the-art AI systems have primarily relied on black-box access, in which auditors can only query the system and observe its outputs. However, white-box access to the system's inner workings (e.g., weights, activations, gradients) allows an auditor to perform stronger attacks, more thoroughly interpret models, and conduct fine-tuning. Meanwhile, outside-the-box access to training and d"},"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":"2401.14446","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CY","submitted_at":"2024-01-25T18:58:05Z","cross_cats_sorted":["cs.AI","cs.CR"],"title_canon_sha256":"44fbdf5480ea790a087350d4ff47b314c7e87cec0e90cda19c288d4b49c5aecd","abstract_canon_sha256":"8f3bc379ae21c9b04c6704d42d2b8631ec8a5b6e748788469185b832fa1de45d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:28:58.796880Z","signature_b64":"73rLOgbNsyuk+KC8wWZ1wVDz48dFekX+XX0xE3C8YUCItoyD0BK+T8POhxcWBO7rOskXdEY+U6Sl7UKxif6ZAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aa0c40680ce5e6be6ca3c60492942db6971197ca337366b5312d17526301222e","last_reissued_at":"2026-07-05T08:28:58.796412Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:28:58.796412Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Black-Box Access is Insufficient for Rigorous AI Audits","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CR"],"primary_cat":"cs.CY","authors_text":"Alan Chan, Andreas Haupt, Benjamin Bucknall, Carson Ezell, Charlotte Siegmann, David Bau, David Krueger, Dylan Hadfield-Menell, J\\'er\\'emy Scheurer, Kevin Wei, Lee Sharkey, Marius Hobbhahn, Marvin Von Hagen, Max Tegmark, Michael Gerovitch, Noam Kolt, Qinyi Sun, Satyapriya Krishna, Silas Alberti, Stephen Casper, Taylor Lynn Curtis","submitted_at":"2024-01-25T18:58:05Z","abstract_excerpt":"External audits of AI systems are increasingly recognized as a key mechanism for AI governance. The effectiveness of an audit, however, depends on the degree of access granted to auditors. Recent audits of state-of-the-art AI systems have primarily relied on black-box access, in which auditors can only query the system and observe its outputs. However, white-box access to the system's inner workings (e.g., weights, activations, gradients) allows an auditor to perform stronger attacks, more thoroughly interpret models, and conduct fine-tuning. Meanwhile, outside-the-box access to training and d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.14446","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/2401.14446/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":"2401.14446","created_at":"2026-07-05T08:28:58.796478+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.14446v3","created_at":"2026-07-05T08:28:58.796478+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.14446","created_at":"2026-07-05T08:28:58.796478+00:00"},{"alias_kind":"pith_short_12","alias_value":"VIGEA2AM4XTL","created_at":"2026-07-05T08:28:58.796478+00:00"},{"alias_kind":"pith_short_16","alias_value":"VIGEA2AM4XTL43FD","created_at":"2026-07-05T08:28:58.796478+00:00"},{"alias_kind":"pith_short_8","alias_value":"VIGEA2AM","created_at":"2026-07-05T08:28:58.796478+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.08044","citing_title":"When Behavioral Safety Evaluation Fails: A Representation-Level Perspective","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21816","citing_title":"Barriers to Evidence in AI-Related Cases and the Privatization of Proof","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2","json":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2.json","graph_json":"https://pith.science/api/pith-number/VIGEA2AM4XTL43FDYYCJFFBNW2/graph.json","events_json":"https://pith.science/api/pith-number/VIGEA2AM4XTL43FDYYCJFFBNW2/events.json","paper":"https://pith.science/paper/VIGEA2AM"},"agent_actions":{"view_html":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2","download_json":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2.json","view_paper":"https://pith.science/paper/VIGEA2AM","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.14446&json=true","fetch_graph":"https://pith.science/api/pith-number/VIGEA2AM4XTL43FDYYCJFFBNW2/graph.json","fetch_events":"https://pith.science/api/pith-number/VIGEA2AM4XTL43FDYYCJFFBNW2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2/action/storage_attestation","attest_author":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2/action/author_attestation","sign_citation":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2/action/citation_signature","submit_replication":"https://pith.science/pith/VIGEA2AM4XTL43FDYYCJFFBNW2/action/replication_record"}},"created_at":"2026-07-05T08:28:58.796478+00:00","updated_at":"2026-07-05T08:28:58.796478+00:00"}