{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:OBTBAALAPN7UBM5SAZ2PGBQ2Q3","short_pith_number":"pith:OBTBAALA","schema_version":"1.0","canonical_sha256":"70661001607b7f40b3b20674f3061a86ce7efb27f2c044724e17721f11a2b7dc","source":{"kind":"arxiv","id":"2605.28137","version":1},"attestation_state":"computed","paper":{"title":"No Safe Dose: How Training Data Drives Unsafe Image Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Felix Friedrich, Kristian Kersting, Lukas Helff, Niharika Hegde, Patrick Schramowski","submitted_at":"2026-05-27T08:21:38Z","abstract_excerpt":"Text-to-image models trained on large-scale data often inevitably ingest unsafe content. While some people observe input-output amplifications, it remains unclear whether and how training data composition directly drives model output safety or by other factors. We shed light on this question by isolating this variable: we train the same text-to-image model on datasets that differ \\emph{only} in their fraction of unsafe images (0\\% to 9.6\\%), across several dataset scales (100K to 8M). Then we generate images with the resulting models, and evaluate them with four independent safety classifiers."},"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":"2605.28137","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-27T08:21:38Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"8cfcaafc4df3b31db67eab980569e7ad41ca9289a28a91a17d7a647d92c88edc","abstract_canon_sha256":"1cabddc1ada2bdbaaf3d8a9550fe0578c14fdc0f8c61661a798ed6ff9d7582b4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:05:00.152898Z","signature_b64":"YhvT5C21x0VL3vONxhaKZX2xzts0Sxb83chIZQ6V9Be9qTr8Iq6EVzUxp84chyR2Uc0d7jRjCYgzqSTttt81AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"70661001607b7f40b3b20674f3061a86ce7efb27f2c044724e17721f11a2b7dc","last_reissued_at":"2026-05-28T01:05:00.152480Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:05:00.152480Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"No Safe Dose: How Training Data Drives Unsafe Image Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Felix Friedrich, Kristian Kersting, Lukas Helff, Niharika Hegde, Patrick Schramowski","submitted_at":"2026-05-27T08:21:38Z","abstract_excerpt":"Text-to-image models trained on large-scale data often inevitably ingest unsafe content. While some people observe input-output amplifications, it remains unclear whether and how training data composition directly drives model output safety or by other factors. We shed light on this question by isolating this variable: we train the same text-to-image model on datasets that differ \\emph{only} in their fraction of unsafe images (0\\% to 9.6\\%), across several dataset scales (100K to 8M). Then we generate images with the resulting models, and evaluate them with four independent safety classifiers."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28137","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/2605.28137/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":"2605.28137","created_at":"2026-05-28T01:05:00.152548+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28137v1","created_at":"2026-05-28T01:05:00.152548+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28137","created_at":"2026-05-28T01:05:00.152548+00:00"},{"alias_kind":"pith_short_12","alias_value":"OBTBAALAPN7U","created_at":"2026-05-28T01:05:00.152548+00:00"},{"alias_kind":"pith_short_16","alias_value":"OBTBAALAPN7UBM5S","created_at":"2026-05-28T01:05:00.152548+00:00"},{"alias_kind":"pith_short_8","alias_value":"OBTBAALA","created_at":"2026-05-28T01:05:00.152548+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/OBTBAALAPN7UBM5SAZ2PGBQ2Q3","json":"https://pith.science/pith/OBTBAALAPN7UBM5SAZ2PGBQ2Q3.json","graph_json":"https://pith.science/api/pith-number/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/graph.json","events_json":"https://pith.science/api/pith-number/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/events.json","paper":"https://pith.science/paper/OBTBAALA"},"agent_actions":{"view_html":"https://pith.science/pith/OBTBAALAPN7UBM5SAZ2PGBQ2Q3","download_json":"https://pith.science/pith/OBTBAALAPN7UBM5SAZ2PGBQ2Q3.json","view_paper":"https://pith.science/paper/OBTBAALA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28137&json=true","fetch_graph":"https://pith.science/api/pith-number/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/graph.json","fetch_events":"https://pith.science/api/pith-number/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/action/storage_attestation","attest_author":"https://pith.science/pith/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/action/author_attestation","sign_citation":"https://pith.science/pith/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/action/citation_signature","submit_replication":"https://pith.science/pith/OBTBAALAPN7UBM5SAZ2PGBQ2Q3/action/replication_record"}},"created_at":"2026-05-28T01:05:00.152548+00:00","updated_at":"2026-05-28T01:05:00.152548+00:00"}