{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PGRGIJNWPK32XSIOXD2WPYI7VP","short_pith_number":"pith:PGRGIJNW","schema_version":"1.0","canonical_sha256":"79a26425b67ab7abc90eb8f567e11fabd57628f5ddd91c9b87b7bbf5bf4ac9b7","source":{"kind":"arxiv","id":"2404.00473","version":1},"attestation_state":"computed","paper":{"title":"Privacy Backdoors: Stealing Data with Corrupted Pretrained Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Florian Tram\\`er, Shanglun Feng","submitted_at":"2024-03-30T20:43:53Z","abstract_excerpt":"Practitioners commonly download pretrained machine learning models from open repositories and finetune them to fit specific applications. We show that this practice introduces a new risk of privacy backdoors. By tampering with a pretrained model's weights, an attacker can fully compromise the privacy of the finetuning data. We show how to build privacy backdoors for a variety of models, including transformers, which enable an attacker to reconstruct individual finetuning samples, with a guaranteed success! We further show that backdoored models allow for tight privacy attacks on models trained"},"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":"2404.00473","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2024-03-30T20:43:53Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"e56a06e027b014c9acc2837736618702868367de27b9048c1cf79b206508103d","abstract_canon_sha256":"efd90cabe701693abc3bbc687bacd848c8989c15403541ce6bb855f2460c73b6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:02:33.873954Z","signature_b64":"DY+XithoBAk/sjsrpFnSF1NEhCGNAXCQIFbVOuD1g1fGt+q2mCTnWVrq/KiWCM8ryCsGcClYvKVQ17DuWp8ABA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"79a26425b67ab7abc90eb8f567e11fabd57628f5ddd91c9b87b7bbf5bf4ac9b7","last_reissued_at":"2026-07-05T08:02:33.873447Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:02:33.873447Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Privacy Backdoors: Stealing Data with Corrupted Pretrained Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Florian Tram\\`er, Shanglun Feng","submitted_at":"2024-03-30T20:43:53Z","abstract_excerpt":"Practitioners commonly download pretrained machine learning models from open repositories and finetune them to fit specific applications. We show that this practice introduces a new risk of privacy backdoors. By tampering with a pretrained model's weights, an attacker can fully compromise the privacy of the finetuning data. We show how to build privacy backdoors for a variety of models, including transformers, which enable an attacker to reconstruct individual finetuning samples, with a guaranteed success! We further show that backdoored models allow for tight privacy attacks on models trained"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.00473","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/2404.00473/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":"2404.00473","created_at":"2026-07-05T08:02:33.873510+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.00473v1","created_at":"2026-07-05T08:02:33.873510+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.00473","created_at":"2026-07-05T08:02:33.873510+00:00"},{"alias_kind":"pith_short_12","alias_value":"PGRGIJNWPK32","created_at":"2026-07-05T08:02:33.873510+00:00"},{"alias_kind":"pith_short_16","alias_value":"PGRGIJNWPK32XSIO","created_at":"2026-07-05T08:02:33.873510+00:00"},{"alias_kind":"pith_short_8","alias_value":"PGRGIJNW","created_at":"2026-07-05T08:02:33.873510+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.31208","citing_title":"Probing Memorization of Tabular In-Context Learning","ref_index":102,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP","json":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP.json","graph_json":"https://pith.science/api/pith-number/PGRGIJNWPK32XSIOXD2WPYI7VP/graph.json","events_json":"https://pith.science/api/pith-number/PGRGIJNWPK32XSIOXD2WPYI7VP/events.json","paper":"https://pith.science/paper/PGRGIJNW"},"agent_actions":{"view_html":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP","download_json":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP.json","view_paper":"https://pith.science/paper/PGRGIJNW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.00473&json=true","fetch_graph":"https://pith.science/api/pith-number/PGRGIJNWPK32XSIOXD2WPYI7VP/graph.json","fetch_events":"https://pith.science/api/pith-number/PGRGIJNWPK32XSIOXD2WPYI7VP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP/action/storage_attestation","attest_author":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP/action/author_attestation","sign_citation":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP/action/citation_signature","submit_replication":"https://pith.science/pith/PGRGIJNWPK32XSIOXD2WPYI7VP/action/replication_record"}},"created_at":"2026-07-05T08:02:33.873510+00:00","updated_at":"2026-07-05T08:02:33.873510+00:00"}