{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:MANLWPXVUIGOKFYEXOE2YGEHAD","short_pith_number":"pith:MANLWPXV","schema_version":"1.0","canonical_sha256":"601abb3ef5a20ce51704bb89ac188700c88a413861e67d692e7eb8875e52c07f","source":{"kind":"arxiv","id":"2310.06816","version":1},"attestation_state":"computed","paper":{"title":"Text Embeddings Reveal (Almost) As Much As Text","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexander M. Rush, John X. Morris, Vitaly Shmatikov, Volodymyr Kuleshov","submitted_at":"2023-10-10T17:39:03Z","abstract_excerpt":"How much private information do text embeddings reveal about the original text? We investigate the problem of embedding \\textit{inversion}, reconstructing the full text represented in dense text embeddings. We frame the problem as controlled generation: generating text that, when reembedded, is close to a fixed point in latent space. We find that although a na\\\"ive model conditioned on the embedding performs poorly, a multi-step method that iteratively corrects and re-embeds text is able to recover $92\\%$ of $32\\text{-token}$ text inputs exactly. We train our model to decode text embeddings fr"},"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":"2310.06816","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-10-10T17:39:03Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"67fe7dbbae993867167c4d6f90bd66ed5dcd5f19d340f10d7e6fcd6bb53d99ba","abstract_canon_sha256":"80e3970f552aaa2ca38bb63f959515a9585c2a90798f8d8448e5f2a2e659633b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:59:24.448865Z","signature_b64":"4ww95DrnRHiLkkn+3tyDPJGlZzriXIzVdX9WX7Yl6CDmBcn5Nhvxi0S7j3Y5aluLkuvr8a2YLwJE3zcfpjonBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"601abb3ef5a20ce51704bb89ac188700c88a413861e67d692e7eb8875e52c07f","last_reissued_at":"2026-07-05T06:59:24.448429Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:59:24.448429Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Text Embeddings Reveal (Almost) As Much As Text","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Alexander M. Rush, John X. Morris, Vitaly Shmatikov, Volodymyr Kuleshov","submitted_at":"2023-10-10T17:39:03Z","abstract_excerpt":"How much private information do text embeddings reveal about the original text? We investigate the problem of embedding \\textit{inversion}, reconstructing the full text represented in dense text embeddings. We frame the problem as controlled generation: generating text that, when reembedded, is close to a fixed point in latent space. We find that although a na\\\"ive model conditioned on the embedding performs poorly, a multi-step method that iteratively corrects and re-embeds text is able to recover $92\\%$ of $32\\text{-token}$ text inputs exactly. We train our model to decode text embeddings fr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.06816","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/2310.06816/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":"2310.06816","created_at":"2026-07-05T06:59:24.448487+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.06816v1","created_at":"2026-07-05T06:59:24.448487+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.06816","created_at":"2026-07-05T06:59:24.448487+00:00"},{"alias_kind":"pith_short_12","alias_value":"MANLWPXVUIGO","created_at":"2026-07-05T06:59:24.448487+00:00"},{"alias_kind":"pith_short_16","alias_value":"MANLWPXVUIGOKFYE","created_at":"2026-07-05T06:59:24.448487+00:00"},{"alias_kind":"pith_short_8","alias_value":"MANLWPXV","created_at":"2026-07-05T06:59:24.448487+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10173","citing_title":"Local Is Not a Sufficient Privacy Boundary: Governing OS-Integrated On-Device AI","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19847","citing_title":"Auditing Privacy in Multi-Tenant RAG under Account Collusion","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2603.15842","citing_title":"Informationally Compressive Anonymization: Non-Degrading Sensitive Input Protection for Privacy-Preserving Supervised Machine Learning","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08693","citing_title":"Towards Generalizable Representations of Mathematical Strategies","ref_index":46,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD","json":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD.json","graph_json":"https://pith.science/api/pith-number/MANLWPXVUIGOKFYEXOE2YGEHAD/graph.json","events_json":"https://pith.science/api/pith-number/MANLWPXVUIGOKFYEXOE2YGEHAD/events.json","paper":"https://pith.science/paper/MANLWPXV"},"agent_actions":{"view_html":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD","download_json":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD.json","view_paper":"https://pith.science/paper/MANLWPXV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.06816&json=true","fetch_graph":"https://pith.science/api/pith-number/MANLWPXVUIGOKFYEXOE2YGEHAD/graph.json","fetch_events":"https://pith.science/api/pith-number/MANLWPXVUIGOKFYEXOE2YGEHAD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD/action/storage_attestation","attest_author":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD/action/author_attestation","sign_citation":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD/action/citation_signature","submit_replication":"https://pith.science/pith/MANLWPXVUIGOKFYEXOE2YGEHAD/action/replication_record"}},"created_at":"2026-07-05T06:59:24.448487+00:00","updated_at":"2026-07-05T06:59:24.448487+00:00"}