{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SUDNVRJXYPEWMY7LUGANHCHAKU","short_pith_number":"pith:SUDNVRJX","schema_version":"1.0","canonical_sha256":"9506dac537c3c96663eba180d388e05530cf4304a998ff07bc025c7c5e6922db","source":{"kind":"arxiv","id":"2603.27518","version":3},"attestation_state":"computed","paper":{"title":"Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Aligned LLMs encode harmful refusal in one global hidden-state direction but spread over-refusal across separate task-specific subspaces.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Mark Dras, Usman Naseem, Utsav Maskey","submitted_at":"2026-03-29T04:53:40Z","abstract_excerpt":"Aligned language models that are trained to refuse harmful requests also exhibit over-refusal: they decline safe instructions that seemingly resemble harmful instructions. A natural approach is to ablate the global refusal direction, steering the hidden-state vectors away or towards the harmful-refusal examples, but this corrects over-refusal only incidentally while disrupting the broader refusal mechanism. In this work, we analyse the representational geometry of both refusal types to understand why this happens. We show that harmful-refusal directions are task-agnostic and can be captured by"},"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":"2603.27518","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-29T04:53:40Z","cross_cats_sorted":[],"title_canon_sha256":"b7dd9d502d6f0c0f6a781fdf84e1f7fcd69d73c91989503844322f1f8955580e","abstract_canon_sha256":"36a047054b045a16488b2793b3efb48cac8af020700fbefe2d1ca71f664e5fb3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T01:05:08.264951Z","signature_b64":"OxU0numwt5Et+YRByUHDIrj3YwLRDBZcBFRMuRM6UnO3ZtFe0qCJ4ppMvMiSoLjqq482MZiuWpsXZ42Mx26NDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9506dac537c3c96663eba180d388e05530cf4304a998ff07bc025c7c5e6922db","last_reissued_at":"2026-05-29T01:05:08.264221Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T01:05:08.264221Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Aligned LLMs encode harmful refusal in one global hidden-state direction but spread over-refusal across separate task-specific subspaces.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Mark Dras, Usman Naseem, Utsav Maskey","submitted_at":"2026-03-29T04:53:40Z","abstract_excerpt":"Aligned language models that are trained to refuse harmful requests also exhibit over-refusal: they decline safe instructions that seemingly resemble harmful instructions. A natural approach is to ablate the global refusal direction, steering the hidden-state vectors away or towards the harmful-refusal examples, but this corrects over-refusal only incidentally while disrupting the broader refusal mechanism. In this work, we analyse the representational geometry of both refusal types to understand why this happens. We show that harmful-refusal directions are task-agnostic and can be captured by"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"harmful-refusal directions are task-agnostic and can be captured by a single global vector, whereas over-refusal directions are task-dependent: they reside within the benign task-representation clusters, vary across tasks, and span a higher-dimensional subspace.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That linear probes and direction-finding methods on hidden states accurately isolate the causal mechanisms of refusal rather than merely capturing correlational patterns.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Harmful refusal in aligned LLMs is captured by a single task-agnostic vector, but over-refusal directions are task-dependent, reside in benign task clusters, and span higher-dimensional subspaces.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Aligned LLMs encode harmful refusal in one global hidden-state direction but spread over-refusal across separate task-specific subspaces.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b94d0ddb5c9a53735530c111f2ea71b1a68eb1a50e0d1bfa2851fe702c7b5d5e"},"source":{"id":"2603.27518","kind":"arxiv","version":3},"verdict":{"id":"b2a85814-1f52-4a40-a141-35f7580b78e8","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T22:15:18.036540Z","strongest_claim":"harmful-refusal directions are task-agnostic and can be captured by a single global vector, whereas over-refusal directions are task-dependent: they reside within the benign task-representation clusters, vary across tasks, and span a higher-dimensional subspace.","one_line_summary":"Harmful refusal in aligned LLMs is captured by a single task-agnostic vector, but over-refusal directions are task-dependent, reside in benign task clusters, and span higher-dimensional subspaces.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That linear probes and direction-finding methods on hidden states accurately isolate the causal mechanisms of refusal rather than merely capturing correlational patterns.","pith_extraction_headline":"Aligned LLMs encode harmful refusal in one global hidden-state direction but spread over-refusal across separate task-specific subspaces."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.27518/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":"2603.27518","created_at":"2026-05-29T01:05:08.264311+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.27518v3","created_at":"2026-05-29T01:05:08.264311+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.27518","created_at":"2026-05-29T01:05:08.264311+00:00"},{"alias_kind":"pith_short_12","alias_value":"SUDNVRJXYPEW","created_at":"2026-05-29T01:05:08.264311+00:00"},{"alias_kind":"pith_short_16","alias_value":"SUDNVRJXYPEWMY7L","created_at":"2026-05-29T01:05:08.264311+00:00"},{"alias_kind":"pith_short_8","alias_value":"SUDNVRJX","created_at":"2026-05-29T01:05:08.264311+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/SUDNVRJXYPEWMY7LUGANHCHAKU","json":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU.json","graph_json":"https://pith.science/api/pith-number/SUDNVRJXYPEWMY7LUGANHCHAKU/graph.json","events_json":"https://pith.science/api/pith-number/SUDNVRJXYPEWMY7LUGANHCHAKU/events.json","paper":"https://pith.science/paper/SUDNVRJX"},"agent_actions":{"view_html":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU","download_json":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU.json","view_paper":"https://pith.science/paper/SUDNVRJX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.27518&json=true","fetch_graph":"https://pith.science/api/pith-number/SUDNVRJXYPEWMY7LUGANHCHAKU/graph.json","fetch_events":"https://pith.science/api/pith-number/SUDNVRJXYPEWMY7LUGANHCHAKU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/action/storage_attestation","attest_author":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/action/author_attestation","sign_citation":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/action/citation_signature","submit_replication":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/action/replication_record"}},"created_at":"2026-05-29T01:05:08.264311+00:00","updated_at":"2026-05-29T01:05:08.264311+00:00"}