{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:UTPLHNCGY6R3DGUEVS6245ZUUJ","short_pith_number":"pith:UTPLHNCG","schema_version":"1.0","canonical_sha256":"a4deb3b446c7a3b19a84acbdae7734a26d6d27dabcef6d17b5215edc8ff48f16","source":{"kind":"arxiv","id":"2505.04796","version":2},"attestation_state":"computed","paper":{"title":"Robust ML Auditing using Prior Knowledge","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Anne-Marie Kermarrec, Augustin Godinot, Erwan Le Merrer, Gilles Tredan, Jade Garcia Bourr\\'ee, Martijn de Vos, Milos Vujasinovic, Sayan Biswas","submitted_at":"2025-05-07T20:46:48Z","abstract_excerpt":"Among the many technical challenges to enforcing AI regulations, one crucial yet underexplored problem is the risk of audit manipulation. This manipulation occurs when a platform deliberately alters its answers to a regulator to pass an audit without modifying its answers to other users. In this paper, we introduce a novel approach to manipulation-proof auditing by taking into account the auditor's prior knowledge of the task solved by the platform. We first demonstrate that regulators must not rely on public priors (e.g. a public dataset), as platforms could easily fool the auditor in such ca"},"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":"2505.04796","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-07T20:46:48Z","cross_cats_sorted":[],"title_canon_sha256":"f37fdf5747b4decdd91b11aa50608a7e8439b0022e827f79e6288f1f04c66e1b","abstract_canon_sha256":"417f04815f6eb4fa07160be0ad7b1bc188eb3363916f07c1ecf4a7acc5e027ca"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:08:20.193978Z","signature_b64":"cX9WZTAlaiyDbMYp+jDCEwgnd/sLlVUx5akgZsKFWMcoaQ7WlGOqsEtvityaejw2+IabraEYX9BruUWsIvorAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a4deb3b446c7a3b19a84acbdae7734a26d6d27dabcef6d17b5215edc8ff48f16","last_reissued_at":"2026-07-05T11:08:20.193431Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:08:20.193431Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Robust ML Auditing using Prior Knowledge","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Anne-Marie Kermarrec, Augustin Godinot, Erwan Le Merrer, Gilles Tredan, Jade Garcia Bourr\\'ee, Martijn de Vos, Milos Vujasinovic, Sayan Biswas","submitted_at":"2025-05-07T20:46:48Z","abstract_excerpt":"Among the many technical challenges to enforcing AI regulations, one crucial yet underexplored problem is the risk of audit manipulation. This manipulation occurs when a platform deliberately alters its answers to a regulator to pass an audit without modifying its answers to other users. In this paper, we introduce a novel approach to manipulation-proof auditing by taking into account the auditor's prior knowledge of the task solved by the platform. We first demonstrate that regulators must not rely on public priors (e.g. a public dataset), as platforms could easily fool the auditor in such ca"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.04796","kind":"arxiv","version":2},"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/2505.04796/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":"2505.04796","created_at":"2026-07-05T11:08:20.193509+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.04796v2","created_at":"2026-07-05T11:08:20.193509+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.04796","created_at":"2026-07-05T11:08:20.193509+00:00"},{"alias_kind":"pith_short_12","alias_value":"UTPLHNCGY6R3","created_at":"2026-07-05T11:08:20.193509+00:00"},{"alias_kind":"pith_short_16","alias_value":"UTPLHNCGY6R3DGUE","created_at":"2026-07-05T11:08:20.193509+00:00"},{"alias_kind":"pith_short_8","alias_value":"UTPLHNCG","created_at":"2026-07-05T11:08:20.193509+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.03330","citing_title":"FLIPS: Instance-Fingerprinting for LLMs via Pseudo-random Sequences","ref_index":44,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ","json":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ.json","graph_json":"https://pith.science/api/pith-number/UTPLHNCGY6R3DGUEVS6245ZUUJ/graph.json","events_json":"https://pith.science/api/pith-number/UTPLHNCGY6R3DGUEVS6245ZUUJ/events.json","paper":"https://pith.science/paper/UTPLHNCG"},"agent_actions":{"view_html":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ","download_json":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ.json","view_paper":"https://pith.science/paper/UTPLHNCG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.04796&json=true","fetch_graph":"https://pith.science/api/pith-number/UTPLHNCGY6R3DGUEVS6245ZUUJ/graph.json","fetch_events":"https://pith.science/api/pith-number/UTPLHNCGY6R3DGUEVS6245ZUUJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ/action/storage_attestation","attest_author":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ/action/author_attestation","sign_citation":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ/action/citation_signature","submit_replication":"https://pith.science/pith/UTPLHNCGY6R3DGUEVS6245ZUUJ/action/replication_record"}},"created_at":"2026-07-05T11:08:20.193509+00:00","updated_at":"2026-07-05T11:08:20.193509+00:00"}