{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:SUDNVRJXYPEWMY7LUGANHCHAKU","short_pith_number":"pith:SUDNVRJX","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"},"canonical_sha256":"9506dac537c3c96663eba180d388e05530cf4304a998ff07bc025c7c5e6922db","source":{"kind":"arxiv","id":"2603.27518","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.27518","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"arxiv_version","alias_value":"2603.27518v3","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.27518","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"pith_short_12","alias_value":"SUDNVRJXYPEW","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"pith_short_16","alias_value":"SUDNVRJXYPEWMY7L","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"pith_short_8","alias_value":"SUDNVRJX","created_at":"2026-05-29T01:05:08Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:SUDNVRJXYPEWMY7LUGANHCHAKU","target":"record","payload":{"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"},"canonical_sha256":"9506dac537c3c96663eba180d388e05530cf4304a998ff07bc025c7c5e6922db","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"},"source_kind":"arxiv","source_id":"2603.27518","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-29T01:05:08Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"pyQEX9M4nUwZESwosJTU6PounfmzYNogPlyXPVz9HVQ/xZ0X0g9XENT2FOy6StDLd0aN5g5LzWjQsbScwiN4Cg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-10T11:28:57.840929Z"},"content_sha256":"d0b7633b3aac838c510bbb68df9d2321f39c052b9605dd95ed02532a2e370068","schema_version":"1.0","event_id":"sha256:d0b7633b3aac838c510bbb68df9d2321f39c052b9605dd95ed02532a2e370068"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:SUDNVRJXYPEWMY7LUGANHCHAKU","target":"graph","payload":{"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"},"verdict_id":"b2a85814-1f52-4a40-a141-35f7580b78e8"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-29T01:05:08Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0Q3ojAwPCtIjPfako4WNSj1SmInjRspyay/B5My+mveqiEKwgd/fFWWlvTdVz4FjdJbkiejVB2w/ho4SBsuGCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-10T11:28:57.841915Z"},"content_sha256":"3832e3a1d97b9f247b6f42c866fb3db3a77f85f71eb45bae9307ab258ac3e133","schema_version":"1.0","event_id":"sha256:3832e3a1d97b9f247b6f42c866fb3db3a77f85f71eb45bae9307ab258ac3e133"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/bundle.json","state_url":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-06-10T11:28:57Z","links":{"resolver":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU","bundle":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/bundle.json","state":"https://pith.science/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/state.json","well_known_bundle":"https://pith.science/.well-known/pith/SUDNVRJXYPEWMY7LUGANHCHAKU/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:SUDNVRJXYPEWMY7LUGANHCHAKU","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"36a047054b045a16488b2793b3efb48cac8af020700fbefe2d1ca71f664e5fb3","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-29T04:53:40Z","title_canon_sha256":"b7dd9d502d6f0c0f6a781fdf84e1f7fcd69d73c91989503844322f1f8955580e"},"schema_version":"1.0","source":{"id":"2603.27518","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2603.27518","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"arxiv_version","alias_value":"2603.27518v3","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.27518","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"pith_short_12","alias_value":"SUDNVRJXYPEW","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"pith_short_16","alias_value":"SUDNVRJXYPEWMY7L","created_at":"2026-05-29T01:05:08Z"},{"alias_kind":"pith_short_8","alias_value":"SUDNVRJX","created_at":"2026-05-29T01:05:08Z"}],"graph_snapshots":[{"event_id":"sha256:3832e3a1d97b9f247b6f42c866fb3db3a77f85f71eb45bae9307ab258ac3e133","target":"graph","created_at":"2026-05-29T01:05:08Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That linear probes and direction-finding methods on hidden states accurately isolate the causal mechanisms of refusal rather than merely capturing correlational patterns."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Aligned LLMs encode harmful refusal in one global hidden-state direction but spread over-refusal across separate task-specific subspaces."}],"snapshot_sha256":"b94d0ddb5c9a53735530c111f2ea71b1a68eb1a50e0d1bfa2851fe702c7b5d5e"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2603.27518/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"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","authors_text":"Mark Dras, Usman Naseem, Utsav Maskey","cross_cats":[],"headline":"Aligned LLMs encode harmful refusal in one global hidden-state direction but spread over-refusal across separate task-specific subspaces.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-29T04:53:40Z","title":"Over-Refusal and Representation Subspaces: A Mechanistic Analysis of Task-Conditioned Refusal in Aligned LLMs"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.27518","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-14T22:15:18.036540Z","id":"b2a85814-1f52-4a40-a141-35f7580b78e8","model_set":{"reader":"grok-4.3"},"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","pith_extraction_headline":"Aligned LLMs encode harmful refusal in one global hidden-state direction but spread over-refusal across separate task-specific subspaces.","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.","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."}},"verdict_id":"b2a85814-1f52-4a40-a141-35f7580b78e8"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:d0b7633b3aac838c510bbb68df9d2321f39c052b9605dd95ed02532a2e370068","target":"record","created_at":"2026-05-29T01:05:08Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"36a047054b045a16488b2793b3efb48cac8af020700fbefe2d1ca71f664e5fb3","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-03-29T04:53:40Z","title_canon_sha256":"b7dd9d502d6f0c0f6a781fdf84e1f7fcd69d73c91989503844322f1f8955580e"},"schema_version":"1.0","source":{"id":"2603.27518","kind":"arxiv","version":3}},"canonical_sha256":"9506dac537c3c96663eba180d388e05530cf4304a998ff07bc025c7c5e6922db","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"9506dac537c3c96663eba180d388e05530cf4304a998ff07bc025c7c5e6922db","first_computed_at":"2026-05-29T01:05:08.264221Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-29T01:05:08.264221Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"OxU0numwt5Et+YRByUHDIrj3YwLRDBZcBFRMuRM6UnO3ZtFe0qCJ4ppMvMiSoLjqq482MZiuWpsXZ42Mx26NDw==","signature_status":"signed_v1","signed_at":"2026-05-29T01:05:08.264951Z","signed_message":"canonical_sha256_bytes"},"source_id":"2603.27518","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:d0b7633b3aac838c510bbb68df9d2321f39c052b9605dd95ed02532a2e370068","sha256:3832e3a1d97b9f247b6f42c866fb3db3a77f85f71eb45bae9307ab258ac3e133"],"state_sha256":"66fd67b8eaa3bae0b87b7780d8e9a9069de642c7f7aae86303093a34d1bdaec0"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"yjhv/P1wG7aL09lj6XByCTwbHC72Dwdbe32DnAoeFo6czbr6l5TBI3ll37gzQ/HZegmqupzu2/tuk6pVRwFmBA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-10T11:28:57.846868Z","bundle_sha256":"73dbf5e14d75e77e7ccafb8e45bcc3ebbed180e066c36584a8b664ac57391fa4"}}