{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:IPK6L6VPIAR4M6PMCWAKEQMHCG","short_pith_number":"pith:IPK6L6VP","schema_version":"1.0","canonical_sha256":"43d5e5faaf4023c679ec1580a2418711ba8b7a270094418c76f4adaa5787b4cf","source":{"kind":"arxiv","id":"2504.21053","version":1},"attestation_state":"computed","paper":{"title":"NeuRel-Attack: Neuron Relearning for Safety Disalignment in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Changting Lin, Dezhang Kong, Meng Han, Wenpeng Xing, Yi Zhou","submitted_at":"2025-04-29T05:49:35Z","abstract_excerpt":"Safety alignment in large language models (LLMs) is achieved through fine-tuning mechanisms that regulate neuron activations to suppress harmful content. In this work, we propose a novel approach to induce disalignment by identifying and modifying the neurons responsible for safety constraints. Our method consists of three key steps: Neuron Activation Analysis, where we examine activation patterns in response to harmful and harmless prompts to detect neurons that are critical for distinguishing between harmful and harmless inputs; Similarity-Based Neuron Identification, which systematically lo"},"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":"2504.21053","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-04-29T05:49:35Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"6c282888994b5d42747968390830a4037e0b5c68f7755bd18cdd6f5a6460c6c1","abstract_canon_sha256":"8060778b0c72cc688d2d203d46f11f42760da2aef9a95276d52fe7db1e737d3d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:55:55.866443Z","signature_b64":"zLXdZIg9mU/MaZpyUpHLhwWTGdKRZ6993d56gh2SVwxobx5wlyAGbxvnX2kyxBrdUJUFLAGVTYwcxu8yZQeVDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"43d5e5faaf4023c679ec1580a2418711ba8b7a270094418c76f4adaa5787b4cf","last_reissued_at":"2026-07-05T10:55:55.865936Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:55:55.865936Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"NeuRel-Attack: Neuron Relearning for Safety Disalignment in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Changting Lin, Dezhang Kong, Meng Han, Wenpeng Xing, Yi Zhou","submitted_at":"2025-04-29T05:49:35Z","abstract_excerpt":"Safety alignment in large language models (LLMs) is achieved through fine-tuning mechanisms that regulate neuron activations to suppress harmful content. In this work, we propose a novel approach to induce disalignment by identifying and modifying the neurons responsible for safety constraints. Our method consists of three key steps: Neuron Activation Analysis, where we examine activation patterns in response to harmful and harmless prompts to detect neurons that are critical for distinguishing between harmful and harmless inputs; Similarity-Based Neuron Identification, which systematically lo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.21053","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/2504.21053/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":"2504.21053","created_at":"2026-07-05T10:55:55.866000+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.21053v1","created_at":"2026-07-05T10:55:55.866000+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.21053","created_at":"2026-07-05T10:55:55.866000+00:00"},{"alias_kind":"pith_short_12","alias_value":"IPK6L6VPIAR4","created_at":"2026-07-05T10:55:55.866000+00:00"},{"alias_kind":"pith_short_16","alias_value":"IPK6L6VPIAR4M6PM","created_at":"2026-07-05T10:55:55.866000+00:00"},{"alias_kind":"pith_short_8","alias_value":"IPK6L6VP","created_at":"2026-07-05T10:55:55.866000+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2512.12069","citing_title":"Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoring","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12384","citing_title":"Preventing Safety Drift in Large Language Models via Coupled Weight and Activation Constraints","ref_index":53,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG","json":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG.json","graph_json":"https://pith.science/api/pith-number/IPK6L6VPIAR4M6PMCWAKEQMHCG/graph.json","events_json":"https://pith.science/api/pith-number/IPK6L6VPIAR4M6PMCWAKEQMHCG/events.json","paper":"https://pith.science/paper/IPK6L6VP"},"agent_actions":{"view_html":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG","download_json":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG.json","view_paper":"https://pith.science/paper/IPK6L6VP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.21053&json=true","fetch_graph":"https://pith.science/api/pith-number/IPK6L6VPIAR4M6PMCWAKEQMHCG/graph.json","fetch_events":"https://pith.science/api/pith-number/IPK6L6VPIAR4M6PMCWAKEQMHCG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG/action/storage_attestation","attest_author":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG/action/author_attestation","sign_citation":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG/action/citation_signature","submit_replication":"https://pith.science/pith/IPK6L6VPIAR4M6PMCWAKEQMHCG/action/replication_record"}},"created_at":"2026-07-05T10:55:55.866000+00:00","updated_at":"2026-07-05T10:55:55.866000+00:00"}