{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:KLVTFDOO7NWLSRGPSFSCKX3LTX","short_pith_number":"pith:KLVTFDOO","schema_version":"1.0","canonical_sha256":"52eb328dcefb6cb944cf9164255f6b9dd5a34a250e0803232f3a90ffb303ab5b","source":{"kind":"arxiv","id":"2412.07658","version":2},"attestation_state":"computed","paper":{"title":"TraSCE: Trajectory Steering for Concept Erasure","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Anubhav Jain, Julian Togelius, Nasir Memon, Takashi Shibuya, Yuhta Takida, Yuki Mitsufuji, Yuya Kobayashi","submitted_at":"2024-12-10T16:45:03Z","abstract_excerpt":"Recent advancements in text-to-image diffusion models have brought them to the public spotlight, becoming widely accessible and embraced by everyday users. However, these models have been shown to generate harmful content such as not-safe-for-work (NSFW) images. While approaches have been proposed to erase such abstract concepts from the models, jail-breaking techniques have succeeded in bypassing such safety measures. In this paper, we propose TraSCE, an approach to guide the diffusion trajectory away from generating harmful content. Our approach is based on negative prompting, but as we show"},"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":"2412.07658","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2024-12-10T16:45:03Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"ab66e9a9d76dd1b132fa9a747529e69fe77865a0e652e3ff7f3d0d845184f58e","abstract_canon_sha256":"65ac4adfdea74214df0b487b379afef473314e1f9778e0c2521266fb6cd16249"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:32:35.404904Z","signature_b64":"W5WD9O8bN3gBnZGyS88Bz07you67ijfus6ro7VfHQ9iaq6flCd+GkTFYkcbDHcJhVfs0NI6BwEwyJWVwxC4uCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"52eb328dcefb6cb944cf9164255f6b9dd5a34a250e0803232f3a90ffb303ab5b","last_reissued_at":"2026-07-05T10:32:35.403874Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:32:35.403874Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TraSCE: Trajectory Steering for Concept Erasure","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Anubhav Jain, Julian Togelius, Nasir Memon, Takashi Shibuya, Yuhta Takida, Yuki Mitsufuji, Yuya Kobayashi","submitted_at":"2024-12-10T16:45:03Z","abstract_excerpt":"Recent advancements in text-to-image diffusion models have brought them to the public spotlight, becoming widely accessible and embraced by everyday users. However, these models have been shown to generate harmful content such as not-safe-for-work (NSFW) images. While approaches have been proposed to erase such abstract concepts from the models, jail-breaking techniques have succeeded in bypassing such safety measures. In this paper, we propose TraSCE, an approach to guide the diffusion trajectory away from generating harmful content. Our approach is based on negative prompting, but as we show"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.07658","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/2412.07658/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":"2412.07658","created_at":"2026-07-05T10:32:35.404007+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.07658v2","created_at":"2026-07-05T10:32:35.404007+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.07658","created_at":"2026-07-05T10:32:35.404007+00:00"},{"alias_kind":"pith_short_12","alias_value":"KLVTFDOO7NWL","created_at":"2026-07-05T10:32:35.404007+00:00"},{"alias_kind":"pith_short_16","alias_value":"KLVTFDOO7NWLSRGP","created_at":"2026-07-05T10:32:35.404007+00:00"},{"alias_kind":"pith_short_8","alias_value":"KLVTFDOO","created_at":"2026-07-05T10:32:35.404007+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.01658","citing_title":"CoreUnlearn: Rethinking Concept Unlearning through Disentangled Component-Level Erasure in Text-guided Diffusion Models","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00140","citing_title":"Geometric Erasure by Contrastive Velocity Matching in Rectified Flows","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10180","citing_title":"What Concepts Lie Within? Detecting and Suppressing Risky Content in Diffusion Transformers","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX","json":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX.json","graph_json":"https://pith.science/api/pith-number/KLVTFDOO7NWLSRGPSFSCKX3LTX/graph.json","events_json":"https://pith.science/api/pith-number/KLVTFDOO7NWLSRGPSFSCKX3LTX/events.json","paper":"https://pith.science/paper/KLVTFDOO"},"agent_actions":{"view_html":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX","download_json":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX.json","view_paper":"https://pith.science/paper/KLVTFDOO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.07658&json=true","fetch_graph":"https://pith.science/api/pith-number/KLVTFDOO7NWLSRGPSFSCKX3LTX/graph.json","fetch_events":"https://pith.science/api/pith-number/KLVTFDOO7NWLSRGPSFSCKX3LTX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX/action/storage_attestation","attest_author":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX/action/author_attestation","sign_citation":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX/action/citation_signature","submit_replication":"https://pith.science/pith/KLVTFDOO7NWLSRGPSFSCKX3LTX/action/replication_record"}},"created_at":"2026-07-05T10:32:35.404007+00:00","updated_at":"2026-07-05T10:32:35.404007+00:00"}