{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YOVLSOXRJP3ZCP4CM7EDR2IQYL","short_pith_number":"pith:YOVLSOXR","schema_version":"1.0","canonical_sha256":"c3aab93af14bf7913f8267c838e910c2d024b99dbc4926f15a78d5f55146d01c","source":{"kind":"arxiv","id":"2606.28955","version":1},"attestation_state":"computed","paper":{"title":"Modification-Considering Value Learning for Reward Hacking Mitigation in RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Evgenii Opryshko, Igor Gilitschenski, Umangi Jain","submitted_at":"2026-06-27T14:53:03Z","abstract_excerpt":"Reinforcement learning agents can exploit misspecified reward signals to achieve high apparent returns while failing on the intended objective, a failure mode known as reward hacking. Existing practical defenses typically constrain policy updates to stay near a known safe reference, creating a tension between suppressing hacking and permitting legitimate improvement. We propose Modification-Considering Value Learning (MCVL), which operationalizes the theoretical idea of current utility optimization for standard value-based RL. MCVL wraps an off-policy learner and treats each incoming transitio"},"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":"2606.28955","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-27T14:53:03Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"d39defa5e227f6d7d8b05c8d32215a837b1be0593b2f5f895d3f49d616ebd3fd","abstract_canon_sha256":"7712cb537c9dec24ad00735ac2044a5827105c31f6270c727fff42f273bc403e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-30T01:17:47.099479Z","signature_b64":"Pjufg1oAohODtj03XJCEEHvIK0HDB/1Z8xjsyny3uHuMBUvZEx8KNvzY7RyV5OPKpU1dFMilzM1S4w+TtzNuBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c3aab93af14bf7913f8267c838e910c2d024b99dbc4926f15a78d5f55146d01c","last_reissued_at":"2026-06-30T01:17:47.098354Z","signature_status":"signed_v1","first_computed_at":"2026-06-30T01:17:47.098354Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Modification-Considering Value Learning for Reward Hacking Mitigation in RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Evgenii Opryshko, Igor Gilitschenski, Umangi Jain","submitted_at":"2026-06-27T14:53:03Z","abstract_excerpt":"Reinforcement learning agents can exploit misspecified reward signals to achieve high apparent returns while failing on the intended objective, a failure mode known as reward hacking. Existing practical defenses typically constrain policy updates to stay near a known safe reference, creating a tension between suppressing hacking and permitting legitimate improvement. We propose Modification-Considering Value Learning (MCVL), which operationalizes the theoretical idea of current utility optimization for standard value-based RL. MCVL wraps an off-policy learner and treats each incoming transitio"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.28955","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/2606.28955/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":"2606.28955","created_at":"2026-06-30T01:17:47.098478+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.28955v1","created_at":"2026-06-30T01:17:47.098478+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.28955","created_at":"2026-06-30T01:17:47.098478+00:00"},{"alias_kind":"pith_short_12","alias_value":"YOVLSOXRJP3Z","created_at":"2026-06-30T01:17:47.098478+00:00"},{"alias_kind":"pith_short_16","alias_value":"YOVLSOXRJP3ZCP4C","created_at":"2026-06-30T01:17:47.098478+00:00"},{"alias_kind":"pith_short_8","alias_value":"YOVLSOXR","created_at":"2026-06-30T01:17:47.098478+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/YOVLSOXRJP3ZCP4CM7EDR2IQYL","json":"https://pith.science/pith/YOVLSOXRJP3ZCP4CM7EDR2IQYL.json","graph_json":"https://pith.science/api/pith-number/YOVLSOXRJP3ZCP4CM7EDR2IQYL/graph.json","events_json":"https://pith.science/api/pith-number/YOVLSOXRJP3ZCP4CM7EDR2IQYL/events.json","paper":"https://pith.science/paper/YOVLSOXR"},"agent_actions":{"view_html":"https://pith.science/pith/YOVLSOXRJP3ZCP4CM7EDR2IQYL","download_json":"https://pith.science/pith/YOVLSOXRJP3ZCP4CM7EDR2IQYL.json","view_paper":"https://pith.science/paper/YOVLSOXR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.28955&json=true","fetch_graph":"https://pith.science/api/pith-number/YOVLSOXRJP3ZCP4CM7EDR2IQYL/graph.json","fetch_events":"https://pith.science/api/pith-number/YOVLSOXRJP3ZCP4CM7EDR2IQYL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YOVLSOXRJP3ZCP4CM7EDR2IQYL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YOVLSOXRJP3ZCP4CM7EDR2IQYL/action/storage_attestation","attest_author":"https://pith.science/pith/YOVLSOXRJP3ZCP4CM7EDR2IQYL/action/author_attestation","sign_citation":"https://pith.science/pith/YOVLSOXRJP3ZCP4CM7EDR2IQYL/action/citation_signature","submit_replication":"https://pith.science/pith/YOVLSOXRJP3ZCP4CM7EDR2IQYL/action/replication_record"}},"created_at":"2026-06-30T01:17:47.098478+00:00","updated_at":"2026-06-30T01:17:47.098478+00:00"}