{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:COFYDDMXGPWBKT4S6YFB4GVEKA","short_pith_number":"pith:COFYDDMX","schema_version":"1.0","canonical_sha256":"138b818d9733ec154f92f60a1e1aa450091211287af18650d843bf4f6eac9155","source":{"kind":"arxiv","id":"2606.25750","version":1},"attestation_state":"computed","paper":{"title":"RAS: Measuring LLM Safety Through Refusal Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Chang-Chieh Huang, Chia-Mu Yu, Wei-Bin Lee, Yan-Lun Chen","submitted_at":"2026-06-24T12:19:40Z","abstract_excerpt":"Safety evaluation of large language models (LLMs) is commonly performed by querying models with unsafe or jailbreak prompts and judging whether their outputs violate a safety policy. Although useful, output-level evaluation is expensive, sensitive to judge choice, and easily tied to fixed question banks. We propose **SafeVec**, a white-box evaluation procedure that measures safety from internal representations rather than generated answers. **SafeVec** first extracts layer-wise refusal directions from a safety-aligned reference model, then selects stable layer windows where safe and unsafe beh"},"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":"2606.25750","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2026-06-24T12:19:40Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"e0d5c0a3a5e46a9c3d32b32c927a4efb8f2c852958e2957ec807eba80ab70da0","abstract_canon_sha256":"4ae45f13bfc8df7779e26b5cfdcf9daaa20d314e0233ec125400969113b9ad7e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-25T01:18:14.270464Z","signature_b64":"ufUXpDuO4rj+dGFQwAAmNGPp4ZTSJVYxZbIN/H6G44CQ6ylvA3rx69abkZ8Uhuz5A9VWkJlUC6BrnZ2YcQTbDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"138b818d9733ec154f92f60a1e1aa450091211287af18650d843bf4f6eac9155","last_reissued_at":"2026-06-25T01:18:14.270081Z","signature_status":"signed_v1","first_computed_at":"2026-06-25T01:18:14.270081Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RAS: Measuring LLM Safety Through Refusal Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CR","authors_text":"Chang-Chieh Huang, Chia-Mu Yu, Wei-Bin Lee, Yan-Lun Chen","submitted_at":"2026-06-24T12:19:40Z","abstract_excerpt":"Safety evaluation of large language models (LLMs) is commonly performed by querying models with unsafe or jailbreak prompts and judging whether their outputs violate a safety policy. Although useful, output-level evaluation is expensive, sensitive to judge choice, and easily tied to fixed question banks. We propose **SafeVec**, a white-box evaluation procedure that measures safety from internal representations rather than generated answers. **SafeVec** first extracts layer-wise refusal directions from a safety-aligned reference model, then selects stable layer windows where safe and unsafe beh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.25750","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/2606.25750/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":"2606.25750","created_at":"2026-06-25T01:18:14.270142+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.25750v1","created_at":"2026-06-25T01:18:14.270142+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.25750","created_at":"2026-06-25T01:18:14.270142+00:00"},{"alias_kind":"pith_short_12","alias_value":"COFYDDMXGPWB","created_at":"2026-06-25T01:18:14.270142+00:00"},{"alias_kind":"pith_short_16","alias_value":"COFYDDMXGPWBKT4S","created_at":"2026-06-25T01:18:14.270142+00:00"},{"alias_kind":"pith_short_8","alias_value":"COFYDDMX","created_at":"2026-06-25T01:18:14.270142+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/COFYDDMXGPWBKT4S6YFB4GVEKA","json":"https://pith.science/pith/COFYDDMXGPWBKT4S6YFB4GVEKA.json","graph_json":"https://pith.science/api/pith-number/COFYDDMXGPWBKT4S6YFB4GVEKA/graph.json","events_json":"https://pith.science/api/pith-number/COFYDDMXGPWBKT4S6YFB4GVEKA/events.json","paper":"https://pith.science/paper/COFYDDMX"},"agent_actions":{"view_html":"https://pith.science/pith/COFYDDMXGPWBKT4S6YFB4GVEKA","download_json":"https://pith.science/pith/COFYDDMXGPWBKT4S6YFB4GVEKA.json","view_paper":"https://pith.science/paper/COFYDDMX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.25750&json=true","fetch_graph":"https://pith.science/api/pith-number/COFYDDMXGPWBKT4S6YFB4GVEKA/graph.json","fetch_events":"https://pith.science/api/pith-number/COFYDDMXGPWBKT4S6YFB4GVEKA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/COFYDDMXGPWBKT4S6YFB4GVEKA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/COFYDDMXGPWBKT4S6YFB4GVEKA/action/storage_attestation","attest_author":"https://pith.science/pith/COFYDDMXGPWBKT4S6YFB4GVEKA/action/author_attestation","sign_citation":"https://pith.science/pith/COFYDDMXGPWBKT4S6YFB4GVEKA/action/citation_signature","submit_replication":"https://pith.science/pith/COFYDDMXGPWBKT4S6YFB4GVEKA/action/replication_record"}},"created_at":"2026-06-25T01:18:14.270142+00:00","updated_at":"2026-06-25T01:18:14.270142+00:00"}