{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:HN3NFV534PGOV3MGDLRORF62CJ","short_pith_number":"pith:HN3NFV53","schema_version":"1.0","canonical_sha256":"3b76d2d7bbe3cceaed861ae2e897da12569bd6a5f241e25a5b9eaf6fc1416247","source":{"kind":"arxiv","id":"2406.11801","version":2},"attestation_state":"computed","paper":{"title":"Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and Activations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Rima Hazra, Sayan Layek, Somnath Banerjee, Soujanya Poria","submitted_at":"2024-06-17T17:48:13Z","abstract_excerpt":"Ensuring the safe alignment of large language models (LLMs) with human values is critical as they become integral to applications like translation and question answering. Current alignment methods struggle with dynamic user intentions and complex objectives, making models vulnerable to generating harmful content. We propose Safety Arithmetic, a training-free framework enhancing LLM safety across different scenarios: Base models, Supervised fine-tuned models (SFT), and Edited models. Safety Arithmetic involves Harm Direction Removal to avoid harmful content and Safety Alignment to promote safe "},"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":"2406.11801","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-06-17T17:48:13Z","cross_cats_sorted":[],"title_canon_sha256":"d4cc6b0df2320837bab151b5232b5f69458e864d975e17af8558986a3fe378bd","abstract_canon_sha256":"57d7648b1703deb87253cd8ad639530b32f8334134af5c248842cf9e43b0182b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:27:02.263810Z","signature_b64":"4DRHtD9bIc/ujRY7jzyCwmqyu3xPtqyVa7jmrENj6OdgWALVB2NEhFubWyYrKOn8K0+voqeRsJqWrHXVsgwEAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3b76d2d7bbe3cceaed861ae2e897da12569bd6a5f241e25a5b9eaf6fc1416247","last_reissued_at":"2026-07-05T09:27:02.263301Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:27:02.263301Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and Activations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Rima Hazra, Sayan Layek, Somnath Banerjee, Soujanya Poria","submitted_at":"2024-06-17T17:48:13Z","abstract_excerpt":"Ensuring the safe alignment of large language models (LLMs) with human values is critical as they become integral to applications like translation and question answering. Current alignment methods struggle with dynamic user intentions and complex objectives, making models vulnerable to generating harmful content. We propose Safety Arithmetic, a training-free framework enhancing LLM safety across different scenarios: Base models, Supervised fine-tuned models (SFT), and Edited models. Safety Arithmetic involves Harm Direction Removal to avoid harmful content and Safety Alignment to promote safe "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.11801","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/2406.11801/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":"2406.11801","created_at":"2026-07-05T09:27:02.263382+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.11801v2","created_at":"2026-07-05T09:27:02.263382+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.11801","created_at":"2026-07-05T09:27:02.263382+00:00"},{"alias_kind":"pith_short_12","alias_value":"HN3NFV534PGO","created_at":"2026-07-05T09:27:02.263382+00:00"},{"alias_kind":"pith_short_16","alias_value":"HN3NFV534PGOV3MG","created_at":"2026-07-05T09:27:02.263382+00:00"},{"alias_kind":"pith_short_8","alias_value":"HN3NFV53","created_at":"2026-07-05T09:27:02.263382+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2408.07666","citing_title":"Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunities","ref_index":68,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ","json":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ.json","graph_json":"https://pith.science/api/pith-number/HN3NFV534PGOV3MGDLRORF62CJ/graph.json","events_json":"https://pith.science/api/pith-number/HN3NFV534PGOV3MGDLRORF62CJ/events.json","paper":"https://pith.science/paper/HN3NFV53"},"agent_actions":{"view_html":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ","download_json":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ.json","view_paper":"https://pith.science/paper/HN3NFV53","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.11801&json=true","fetch_graph":"https://pith.science/api/pith-number/HN3NFV534PGOV3MGDLRORF62CJ/graph.json","fetch_events":"https://pith.science/api/pith-number/HN3NFV534PGOV3MGDLRORF62CJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ/action/storage_attestation","attest_author":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ/action/author_attestation","sign_citation":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ/action/citation_signature","submit_replication":"https://pith.science/pith/HN3NFV534PGOV3MGDLRORF62CJ/action/replication_record"}},"created_at":"2026-07-05T09:27:02.263382+00:00","updated_at":"2026-07-05T09:27:02.263382+00:00"}