{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:X4QILAKKAXVHPJFTH74MLXCYJU","short_pith_number":"pith:X4QILAKK","schema_version":"1.0","canonical_sha256":"bf2085814a05ea77a4b33ff8c5dc584d2a29f67ecf194df91f8de69fcbdeffb9","source":{"kind":"arxiv","id":"2511.19558","version":2},"attestation_state":"computed","paper":{"title":"SPQR: A Multi-Dimensional Benchmark for Safety Alignment under Benign Model Adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.LG"],"primary_cat":"cs.CR","authors_text":"Fahad Shamshad, Fahkri Karray, Karthik Nandakumar, Mohammed Talha Alam, Nada Saadi, Nils Lukas, Samuele Poppi","submitted_at":"2025-11-24T14:46:20Z","abstract_excerpt":"Text-to-image diffusion models can emit copyrighted, unsafe, or private content. Safety alignment aims to suppress specific concepts, yet evaluations seldom test whether safety persists under benign downstream fine-tuning routinely applied after deployment (e.g., LoRA personalization, style/domain adapters). We study the stability of current safety methods under benign fine-tuning and observe frequent breakdowns. As true safety alignment must withstand even benign post-deployment adaptations, we introduce the SPQR benchmark (Safety, Prompt adherence, Quality, and Robustness). SPQR is a single-"},"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":"2511.19558","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-11-24T14:46:20Z","cross_cats_sorted":["cs.AI","cs.CV","cs.LG"],"title_canon_sha256":"88c6b6e31b2bdbe264d4aff3be051e729540c6b38888974d307e9b64756922b7","abstract_canon_sha256":"83877307a96d08dbde8b89494eac4aaa4369e6b506db57212466cb38258c4bd5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-14T01:20:48.841863Z","signature_b64":"FuZlfSGz48XtFzAReJCFFSQpdPeoszUZyzxYE2wvH1MOp81Rw+MWc1dI+iMhhu5cJmk35wY5RR80a5QDK1m3CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bf2085814a05ea77a4b33ff8c5dc584d2a29f67ecf194df91f8de69fcbdeffb9","last_reissued_at":"2026-07-14T01:20:48.840965Z","signature_status":"signed_v1","first_computed_at":"2026-07-14T01:20:48.840965Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SPQR: A Multi-Dimensional Benchmark for Safety Alignment under Benign Model Adaptation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.LG"],"primary_cat":"cs.CR","authors_text":"Fahad Shamshad, Fahkri Karray, Karthik Nandakumar, Mohammed Talha Alam, Nada Saadi, Nils Lukas, Samuele Poppi","submitted_at":"2025-11-24T14:46:20Z","abstract_excerpt":"Text-to-image diffusion models can emit copyrighted, unsafe, or private content. Safety alignment aims to suppress specific concepts, yet evaluations seldom test whether safety persists under benign downstream fine-tuning routinely applied after deployment (e.g., LoRA personalization, style/domain adapters). We study the stability of current safety methods under benign fine-tuning and observe frequent breakdowns. As true safety alignment must withstand even benign post-deployment adaptations, we introduce the SPQR benchmark (Safety, Prompt adherence, Quality, and Robustness). SPQR is a single-"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2511.19558","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/2511.19558/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":"2511.19558","created_at":"2026-07-14T01:20:48.841381+00:00"},{"alias_kind":"arxiv_version","alias_value":"2511.19558v2","created_at":"2026-07-14T01:20:48.841381+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.19558","created_at":"2026-07-14T01:20:48.841381+00:00"},{"alias_kind":"pith_short_12","alias_value":"X4QILAKKAXVH","created_at":"2026-07-14T01:20:48.841381+00:00"},{"alias_kind":"pith_short_16","alias_value":"X4QILAKKAXVHPJFT","created_at":"2026-07-14T01:20:48.841381+00:00"},{"alias_kind":"pith_short_8","alias_value":"X4QILAKK","created_at":"2026-07-14T01:20:48.841381+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.18719","citing_title":"SafeDiffusion-R1: Online Reward Steering for Safe Diffusion Post-Training","ref_index":100,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21041","citing_title":"Projected Gradient Unlearning for Text-to-Image Diffusion Models: Defending Against Concept Revival Attacks","ref_index":13,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU","json":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU.json","graph_json":"https://pith.science/api/pith-number/X4QILAKKAXVHPJFTH74MLXCYJU/graph.json","events_json":"https://pith.science/api/pith-number/X4QILAKKAXVHPJFTH74MLXCYJU/events.json","paper":"https://pith.science/paper/X4QILAKK"},"agent_actions":{"view_html":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU","download_json":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU.json","view_paper":"https://pith.science/paper/X4QILAKK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2511.19558&json=true","fetch_graph":"https://pith.science/api/pith-number/X4QILAKKAXVHPJFTH74MLXCYJU/graph.json","fetch_events":"https://pith.science/api/pith-number/X4QILAKKAXVHPJFTH74MLXCYJU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU/action/storage_attestation","attest_author":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU/action/author_attestation","sign_citation":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU/action/citation_signature","submit_replication":"https://pith.science/pith/X4QILAKKAXVHPJFTH74MLXCYJU/action/replication_record"}},"created_at":"2026-07-14T01:20:48.841381+00:00","updated_at":"2026-07-14T01:20:48.841381+00:00"}