{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:XJLW2WGEXTUBDADXCCO5BTUZGK","short_pith_number":"pith:XJLW2WGE","canonical_record":{"source":{"id":"2506.04250","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-01T01:19:37Z","cross_cats_sorted":[],"title_canon_sha256":"0829b3ac5313b4d89d3623c1c92f3f7ae922fa28cb39d6f7a03db8cda6cc2855","abstract_canon_sha256":"58b0cfeb78abe70758779e5a94ded20ee6675e249c32eb5c80690ba9858dbf81"},"schema_version":"1.0"},"canonical_sha256":"ba576d58c4bce8118077109dd0ce9932833ee87ab0f8d1817d8e208e164321b2","source":{"kind":"arxiv","id":"2506.04250","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.04250","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"arxiv_version","alias_value":"2506.04250v1","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.04250","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"pith_short_12","alias_value":"XJLW2WGEXTUB","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"pith_short_16","alias_value":"XJLW2WGEXTUBDADX","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"pith_short_8","alias_value":"XJLW2WGE","created_at":"2026-07-05T11:16:11Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:XJLW2WGEXTUBDADXCCO5BTUZGK","target":"record","payload":{"canonical_record":{"source":{"id":"2506.04250","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-01T01:19:37Z","cross_cats_sorted":[],"title_canon_sha256":"0829b3ac5313b4d89d3623c1c92f3f7ae922fa28cb39d6f7a03db8cda6cc2855","abstract_canon_sha256":"58b0cfeb78abe70758779e5a94ded20ee6675e249c32eb5c80690ba9858dbf81"},"schema_version":"1.0"},"canonical_sha256":"ba576d58c4bce8118077109dd0ce9932833ee87ab0f8d1817d8e208e164321b2","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:16:11.089007Z","signature_b64":"YW6+duo14IqO3LQKNhov13XRrN/7E7mDFDeOihma7pB3+Y2J4nrVeJiLZKs0wr/l3qdil1Y0B3o4QN6YlFy6DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ba576d58c4bce8118077109dd0ce9932833ee87ab0f8d1817d8e208e164321b2","last_reissued_at":"2026-07-05T11:16:11.088499Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:16:11.088499Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2506.04250","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:16:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"2FMdKxFRqol8ZSAzYCHuul4Zqu20CcNkUcXlRuyLSMH2zAKYVRx4lgluSERfmuB6uerY18AZDz/hU/kg8iy8BA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-09T14:37:35.905249Z"},"content_sha256":"fadb7e5ff590f49da0d79948b933a5a9daff79f90f851de556fd982868b7099d","schema_version":"1.0","event_id":"sha256:fadb7e5ff590f49da0d79948b933a5a9daff79f90f851de556fd982868b7099d"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:XJLW2WGEXTUBDADXCCO5BTUZGK","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"SafeSteer: Interpretable Safety Steering with Refusal-Evasion in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Amrita Bhattacharjee, Christopher Parisien, Shaona Ghosh, Yftah Ziser","submitted_at":"2025-06-01T01:19:37Z","abstract_excerpt":"Fine-tuning large language models (LLMs) to adapt to evolving safety policies is costly and impractical. Mechanistic interpretability enables inference-time control through latent activation steering, yet its potential for precise, customizable safety adjustments remains largely untapped. This paper investigates an approach called SafeSteer for guiding the outputs of LLMs by: (i) leveraging category-specific steering vectors for more precise control, (ii) employing a simple, gradient-free unsupervised method to enhance safety steering while preserving text quality, topic relevance, and without"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.04250","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/2506.04250/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":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:16:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"zpr057EgNyEU3CZNeJxKfXqn4EIclp1AMgUKwF3oaeNdEbdXpXKeIhYQ49YBW1mll3C+ER2LKvZ/S0RZRoe1Cw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-09T14:37:35.905761Z"},"content_sha256":"2cdc03426a3c3e71c0ef24950b2795b32b7b592ed49a4f60d1c03189a78b9ef9","schema_version":"1.0","event_id":"sha256:2cdc03426a3c3e71c0ef24950b2795b32b7b592ed49a4f60d1c03189a78b9ef9"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/XJLW2WGEXTUBDADXCCO5BTUZGK/bundle.json","state_url":"https://pith.science/pith/XJLW2WGEXTUBDADXCCO5BTUZGK/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/XJLW2WGEXTUBDADXCCO5BTUZGK/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":"2026-08-09T14:37:35Z","links":{"resolver":"https://pith.science/pith/XJLW2WGEXTUBDADXCCO5BTUZGK","bundle":"https://pith.science/pith/XJLW2WGEXTUBDADXCCO5BTUZGK/bundle.json","state":"https://pith.science/pith/XJLW2WGEXTUBDADXCCO5BTUZGK/state.json","well_known_bundle":"https://pith.science/.well-known/pith/XJLW2WGEXTUBDADXCCO5BTUZGK/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:XJLW2WGEXTUBDADXCCO5BTUZGK","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":"58b0cfeb78abe70758779e5a94ded20ee6675e249c32eb5c80690ba9858dbf81","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-01T01:19:37Z","title_canon_sha256":"0829b3ac5313b4d89d3623c1c92f3f7ae922fa28cb39d6f7a03db8cda6cc2855"},"schema_version":"1.0","source":{"id":"2506.04250","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.04250","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"arxiv_version","alias_value":"2506.04250v1","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.04250","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"pith_short_12","alias_value":"XJLW2WGEXTUB","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"pith_short_16","alias_value":"XJLW2WGEXTUBDADX","created_at":"2026-07-05T11:16:11Z"},{"alias_kind":"pith_short_8","alias_value":"XJLW2WGE","created_at":"2026-07-05T11:16:11Z"}],"graph_snapshots":[{"event_id":"sha256:2cdc03426a3c3e71c0ef24950b2795b32b7b592ed49a4f60d1c03189a78b9ef9","target":"graph","created_at":"2026-07-05T11:16:11Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2506.04250/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Fine-tuning large language models (LLMs) to adapt to evolving safety policies is costly and impractical. Mechanistic interpretability enables inference-time control through latent activation steering, yet its potential for precise, customizable safety adjustments remains largely untapped. This paper investigates an approach called SafeSteer for guiding the outputs of LLMs by: (i) leveraging category-specific steering vectors for more precise control, (ii) employing a simple, gradient-free unsupervised method to enhance safety steering while preserving text quality, topic relevance, and without","authors_text":"Amrita Bhattacharjee, Christopher Parisien, Shaona Ghosh, Yftah Ziser","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-01T01:19:37Z","title":"SafeSteer: Interpretable Safety Steering with Refusal-Evasion in LLMs"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.04250","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:fadb7e5ff590f49da0d79948b933a5a9daff79f90f851de556fd982868b7099d","target":"record","created_at":"2026-07-05T11:16:11Z","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":"58b0cfeb78abe70758779e5a94ded20ee6675e249c32eb5c80690ba9858dbf81","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-01T01:19:37Z","title_canon_sha256":"0829b3ac5313b4d89d3623c1c92f3f7ae922fa28cb39d6f7a03db8cda6cc2855"},"schema_version":"1.0","source":{"id":"2506.04250","kind":"arxiv","version":1}},"canonical_sha256":"ba576d58c4bce8118077109dd0ce9932833ee87ab0f8d1817d8e208e164321b2","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ba576d58c4bce8118077109dd0ce9932833ee87ab0f8d1817d8e208e164321b2","first_computed_at":"2026-07-05T11:16:11.088499Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:16:11.088499Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"YW6+duo14IqO3LQKNhov13XRrN/7E7mDFDeOihma7pB3+Y2J4nrVeJiLZKs0wr/l3qdil1Y0B3o4QN6YlFy6DA==","signature_status":"signed_v1","signed_at":"2026-07-05T11:16:11.089007Z","signed_message":"canonical_sha256_bytes"},"source_id":"2506.04250","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:fadb7e5ff590f49da0d79948b933a5a9daff79f90f851de556fd982868b7099d","sha256:2cdc03426a3c3e71c0ef24950b2795b32b7b592ed49a4f60d1c03189a78b9ef9"],"state_sha256":"b81623590f014d1fa7ea289a44366b31e5cd3c620ef873f7a348252235c046bd"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0GxMeSjbYjPlUxko9GZ3QnKHtRLiiPnZXqn3R2ZztB7ozfYK6zalgwT2GCJVu0A68kmRcRRXnz6TNjd8dzhdBg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-09T14:37:35.911090Z","bundle_sha256":"8e9d500526d51cdb7d09f06ba476b63bd002e5e2e81e24ee9e26cfce6104fea1"}}