{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:G6WN3DOHVZNUGBQFWY5SBGD3UH","short_pith_number":"pith:G6WN3DOH","canonical_record":{"source":{"id":"2605.21225","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-20T14:19:45Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0a8a4ec2718fd16605dd5617706e08c6de26a85959dc69585087e54c4588be29","abstract_canon_sha256":"04fae145ca22c61a0defcd2e368768ff4e0a5b53e57b55c5152c36538a5bf401"},"schema_version":"1.0"},"canonical_sha256":"37acdd8dc7ae5b430605b63b20987ba1e91c0ce2a6bcfcb65e654cc0822c825f","source":{"kind":"arxiv","id":"2605.21225","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.21225","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"arxiv_version","alias_value":"2605.21225v1","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.21225","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"pith_short_12","alias_value":"G6WN3DOHVZNU","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"pith_short_16","alias_value":"G6WN3DOHVZNUGBQF","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"pith_short_8","alias_value":"G6WN3DOH","created_at":"2026-05-21T01:05:44Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:G6WN3DOHVZNUGBQFWY5SBGD3UH","target":"record","payload":{"canonical_record":{"source":{"id":"2605.21225","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-20T14:19:45Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"0a8a4ec2718fd16605dd5617706e08c6de26a85959dc69585087e54c4588be29","abstract_canon_sha256":"04fae145ca22c61a0defcd2e368768ff4e0a5b53e57b55c5152c36538a5bf401"},"schema_version":"1.0"},"canonical_sha256":"37acdd8dc7ae5b430605b63b20987ba1e91c0ce2a6bcfcb65e654cc0822c825f","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T01:05:44.363701Z","signature_b64":"U6Ah9uS38R/NKOS38t9kbI7wj0tKCOQ3CaJrlHPjOjXP4yp/978rIwHgscY+RVnJ0Lv4OlQ6G/BGU6Hv0N6pCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"37acdd8dc7ae5b430605b63b20987ba1e91c0ce2a6bcfcb65e654cc0822c825f","last_reissued_at":"2026-05-21T01:05:44.362943Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T01:05:44.362943Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.21225","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":"2026-05-21T01:05:44Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"o7ZZkJHe+Lmt6HYf9rISPkYTrlEdx3atZ6KhSH/TFlAN32p8zavRG8/q0O+Ho3p1NYmpJdwAkVTB7vuiJxYhDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-08T06:17:24.064228Z"},"content_sha256":"539d0ec3c1ccfce9eacf38399bea100a03cc702c5343487621a1e72a1aecda7a","schema_version":"1.0","event_id":"sha256:539d0ec3c1ccfce9eacf38399bea100a03cc702c5343487621a1e72a1aecda7a"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:G6WN3DOHVZNUGBQFWY5SBGD3UH","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"PREFINE: Preference-Based Implicit Reward and Cost Fine-Tuning for Safety Alignment","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Balaraman Ravindran, Bavish Kulur, Richa Verma, Sanjay Chawla","submitted_at":"2026-05-20T14:19:45Z","abstract_excerpt":"We address the problem of making a pre-trained reinforcement learning (RL) policy safety-aware by incorporating cost constraints without retraining it from scratch. While costs could be numerically encoded, we assume a more general setting is when costs are provided as preferences. Given a reward-optimized policy and a small dataset of preferred (low-cost) and dispreferred (high-cost) trajectories, our goal is to fine-tune the policy to generate low-cost behaviors while retaining high rewards. Unlike standard RLHF in language models, where preferences are defined over responses to the same pro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.21225","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/2605.21225/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":"2026-05-21T01:05:44Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"TNOoOBeIgO3j2lyQvtURq4HNHD70Fhx9pRYmXH8jnIGx82OdnCi7Ye84/hUumUjTdXdlWRltqHifpSyK8kW6BA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-08T06:17:24.064981Z"},"content_sha256":"dc0536ee6cd25cc08037d75ca58b77c9c5236acf3e61b81a9fa984694f131b54","schema_version":"1.0","event_id":"sha256:dc0536ee6cd25cc08037d75ca58b77c9c5236acf3e61b81a9fa984694f131b54"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/G6WN3DOHVZNUGBQFWY5SBGD3UH/bundle.json","state_url":"https://pith.science/pith/G6WN3DOHVZNUGBQFWY5SBGD3UH/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/G6WN3DOHVZNUGBQFWY5SBGD3UH/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":"2026-06-08T06:17:24Z","links":{"resolver":"https://pith.science/pith/G6WN3DOHVZNUGBQFWY5SBGD3UH","bundle":"https://pith.science/pith/G6WN3DOHVZNUGBQFWY5SBGD3UH/bundle.json","state":"https://pith.science/pith/G6WN3DOHVZNUGBQFWY5SBGD3UH/state.json","well_known_bundle":"https://pith.science/.well-known/pith/G6WN3DOHVZNUGBQFWY5SBGD3UH/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:G6WN3DOHVZNUGBQFWY5SBGD3UH","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":"04fae145ca22c61a0defcd2e368768ff4e0a5b53e57b55c5152c36538a5bf401","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-20T14:19:45Z","title_canon_sha256":"0a8a4ec2718fd16605dd5617706e08c6de26a85959dc69585087e54c4588be29"},"schema_version":"1.0","source":{"id":"2605.21225","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.21225","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"arxiv_version","alias_value":"2605.21225v1","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.21225","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"pith_short_12","alias_value":"G6WN3DOHVZNU","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"pith_short_16","alias_value":"G6WN3DOHVZNUGBQF","created_at":"2026-05-21T01:05:44Z"},{"alias_kind":"pith_short_8","alias_value":"G6WN3DOH","created_at":"2026-05-21T01:05:44Z"}],"graph_snapshots":[{"event_id":"sha256:dc0536ee6cd25cc08037d75ca58b77c9c5236acf3e61b81a9fa984694f131b54","target":"graph","created_at":"2026-05-21T01:05:44Z","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/2605.21225/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"We address the problem of making a pre-trained reinforcement learning (RL) policy safety-aware by incorporating cost constraints without retraining it from scratch. While costs could be numerically encoded, we assume a more general setting is when costs are provided as preferences. Given a reward-optimized policy and a small dataset of preferred (low-cost) and dispreferred (high-cost) trajectories, our goal is to fine-tune the policy to generate low-cost behaviors while retaining high rewards. Unlike standard RLHF in language models, where preferences are defined over responses to the same pro","authors_text":"Balaraman Ravindran, Bavish Kulur, Richa Verma, Sanjay Chawla","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-20T14:19:45Z","title":"PREFINE: Preference-Based Implicit Reward and Cost Fine-Tuning for Safety Alignment"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.21225","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":"sha256:539d0ec3c1ccfce9eacf38399bea100a03cc702c5343487621a1e72a1aecda7a","target":"record","created_at":"2026-05-21T01:05:44Z","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":"04fae145ca22c61a0defcd2e368768ff4e0a5b53e57b55c5152c36538a5bf401","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-20T14:19:45Z","title_canon_sha256":"0a8a4ec2718fd16605dd5617706e08c6de26a85959dc69585087e54c4588be29"},"schema_version":"1.0","source":{"id":"2605.21225","kind":"arxiv","version":1}},"canonical_sha256":"37acdd8dc7ae5b430605b63b20987ba1e91c0ce2a6bcfcb65e654cc0822c825f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"37acdd8dc7ae5b430605b63b20987ba1e91c0ce2a6bcfcb65e654cc0822c825f","first_computed_at":"2026-05-21T01:05:44.362943Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-21T01:05:44.362943Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"U6Ah9uS38R/NKOS38t9kbI7wj0tKCOQ3CaJrlHPjOjXP4yp/978rIwHgscY+RVnJ0Lv4OlQ6G/BGU6Hv0N6pCA==","signature_status":"signed_v1","signed_at":"2026-05-21T01:05:44.363701Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.21225","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:539d0ec3c1ccfce9eacf38399bea100a03cc702c5343487621a1e72a1aecda7a","sha256:dc0536ee6cd25cc08037d75ca58b77c9c5236acf3e61b81a9fa984694f131b54"],"state_sha256":"69a71f9afb8d521f01c9d3d7fe6423fd3a43400b550e3a5444288b0a4b0c8ee3"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Z4HwEojohAmhXiElQJpiM94SmuqLv4oy0Dc7C4JiG9lu774R/SJ98D3UFDuxaYFVW5t4hQg6IQD+XyGkl7VwAA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-08T06:17:24.068540Z","bundle_sha256":"991a023762e852f5178c85f6c8c32e377ef37a5ae4b5f9fa4a2f19b5e3b01ee2"}}