{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3QC25UTRE4EN54CRKYTJWBHHJ2","short_pith_number":"pith:3QC25UTR","schema_version":"1.0","canonical_sha256":"dc05aed2712708def05156269b04e74eb887ca1c0be4d13a2c356bbdbe7f6a92","source":{"kind":"arxiv","id":"2606.25442","version":1},"attestation_state":"computed","paper":{"title":"PolicyAlign: Direct Policy-Based Safety Alignment for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chang Wu, Guanjun Jiang, Houcheng Jiang, Junfeng Fang, Kai Tang, Pengyu Cheng, Xiang Wang, Xiaoxi Jiang","submitted_at":"2026-06-24T06:10:33Z","abstract_excerpt":"Safety alignment of large language models (LLMs) typically depends on high-quality supervision data, such as safe demonstrations or preference pairs. However, in real-world deployment, emerging safety requirements are often specified as natural-language policies, while corresponding supervision data may be costly, delayed, or unavailable. This creates a mismatch between rapidly evolving safety policies and conventional data-driven alignment methods. To address this, we propose PolicyAlign, a simple yet effective framework for directly aligning LLMs with safety policies. Given a safety policy, "},"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":"2606.25442","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-24T06:10:33Z","cross_cats_sorted":[],"title_canon_sha256":"d6f573314f0c52ff308ba40f81e2d70b0cb8f3301102e3b0803e5f2855a9f308","abstract_canon_sha256":"b4794242e170127feee3d876cc7a9759b87256cef7fecdc05825178fb9cb36cc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-25T01:18:05.443724Z","signature_b64":"hTSzmsXbwjhJxKOr/bxpRNhy8jfeJfqVcGoG0mVeX59VDNaWntHoun+fjYI0OS5aHQak9/+lQpEDN2xL3bLBAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dc05aed2712708def05156269b04e74eb887ca1c0be4d13a2c356bbdbe7f6a92","last_reissued_at":"2026-06-25T01:18:05.443283Z","signature_status":"signed_v1","first_computed_at":"2026-06-25T01:18:05.443283Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PolicyAlign: Direct Policy-Based Safety Alignment for Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chang Wu, Guanjun Jiang, Houcheng Jiang, Junfeng Fang, Kai Tang, Pengyu Cheng, Xiang Wang, Xiaoxi Jiang","submitted_at":"2026-06-24T06:10:33Z","abstract_excerpt":"Safety alignment of large language models (LLMs) typically depends on high-quality supervision data, such as safe demonstrations or preference pairs. However, in real-world deployment, emerging safety requirements are often specified as natural-language policies, while corresponding supervision data may be costly, delayed, or unavailable. This creates a mismatch between rapidly evolving safety policies and conventional data-driven alignment methods. To address this, we propose PolicyAlign, a simple yet effective framework for directly aligning LLMs with safety policies. Given a safety policy, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.25442","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/2606.25442/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":"2606.25442","created_at":"2026-06-25T01:18:05.443344+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.25442v1","created_at":"2026-06-25T01:18:05.443344+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.25442","created_at":"2026-06-25T01:18:05.443344+00:00"},{"alias_kind":"pith_short_12","alias_value":"3QC25UTRE4EN","created_at":"2026-06-25T01:18:05.443344+00:00"},{"alias_kind":"pith_short_16","alias_value":"3QC25UTRE4EN54CR","created_at":"2026-06-25T01:18:05.443344+00:00"},{"alias_kind":"pith_short_8","alias_value":"3QC25UTR","created_at":"2026-06-25T01:18:05.443344+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/3QC25UTRE4EN54CRKYTJWBHHJ2","json":"https://pith.science/pith/3QC25UTRE4EN54CRKYTJWBHHJ2.json","graph_json":"https://pith.science/api/pith-number/3QC25UTRE4EN54CRKYTJWBHHJ2/graph.json","events_json":"https://pith.science/api/pith-number/3QC25UTRE4EN54CRKYTJWBHHJ2/events.json","paper":"https://pith.science/paper/3QC25UTR"},"agent_actions":{"view_html":"https://pith.science/pith/3QC25UTRE4EN54CRKYTJWBHHJ2","download_json":"https://pith.science/pith/3QC25UTRE4EN54CRKYTJWBHHJ2.json","view_paper":"https://pith.science/paper/3QC25UTR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.25442&json=true","fetch_graph":"https://pith.science/api/pith-number/3QC25UTRE4EN54CRKYTJWBHHJ2/graph.json","fetch_events":"https://pith.science/api/pith-number/3QC25UTRE4EN54CRKYTJWBHHJ2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3QC25UTRE4EN54CRKYTJWBHHJ2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3QC25UTRE4EN54CRKYTJWBHHJ2/action/storage_attestation","attest_author":"https://pith.science/pith/3QC25UTRE4EN54CRKYTJWBHHJ2/action/author_attestation","sign_citation":"https://pith.science/pith/3QC25UTRE4EN54CRKYTJWBHHJ2/action/citation_signature","submit_replication":"https://pith.science/pith/3QC25UTRE4EN54CRKYTJWBHHJ2/action/replication_record"}},"created_at":"2026-06-25T01:18:05.443344+00:00","updated_at":"2026-06-25T01:18:05.443344+00:00"}