{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:WWJFZ7B4WPXRHL3TYTN55IPQVI","short_pith_number":"pith:WWJFZ7B4","schema_version":"1.0","canonical_sha256":"b5925cfc3cb3ef13af73c4dbdea1f0aa247adbc65b8410e6dea99766f8c113b3","source":{"kind":"arxiv","id":"2606.03131","version":1},"attestation_state":"computed","paper":{"title":"HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Caiyue Huang, Mengnan Du, Qiuyang Zhao, Shuang Liu, Xiaorong Chen, Yanguang Liu, Yuxuan Bo","submitted_at":"2026-06-02T04:18:08Z","abstract_excerpt":"Reward models are central to large language model (LLM) alignment, but they remain vulnerable to reward hacking. To evaluate reward-model robustness, we introduce RewardHackBench containing 13 reward-hacking patterns covering real life high-stakes domains and general settings, and we find severe failures on specific subcategories across eight reward models. To mitigate these failures, we propose HARVE, a training-free reward-head editing method for scalar reward models. Instead of fine-tuning the reward model, HARVE identifies a multi-directional hacking subspace from residual stream direction"},"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":"2606.03131","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-02T04:18:08Z","cross_cats_sorted":[],"title_canon_sha256":"16f2d180df6c6ad2c8f011e9ef988e49998e79195da8b9495235d2dcf0e880a2","abstract_canon_sha256":"2b67adc44d2a162c0a701679912227bedca19f8ef41c22dcd67fc7d0a2412cd1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-03T01:05:32.538057Z","signature_b64":"z1W7sOXvCk4x4BxNBZdL+KVuzUep//A2urgS8xQ4tU/BBsE6WKfQ1zSd6Lixf7gIs0XXS/h4ZoLBfItCMhMHDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b5925cfc3cb3ef13af73c4dbdea1f0aa247adbc65b8410e6dea99766f8c113b3","last_reissued_at":"2026-06-03T01:05:32.537698Z","signature_status":"signed_v1","first_computed_at":"2026-06-03T01:05:32.537698Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"HARVE: Hacking-Aware Reward-Head Vector Editing for Robust Reward Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Caiyue Huang, Mengnan Du, Qiuyang Zhao, Shuang Liu, Xiaorong Chen, Yanguang Liu, Yuxuan Bo","submitted_at":"2026-06-02T04:18:08Z","abstract_excerpt":"Reward models are central to large language model (LLM) alignment, but they remain vulnerable to reward hacking. To evaluate reward-model robustness, we introduce RewardHackBench containing 13 reward-hacking patterns covering real life high-stakes domains and general settings, and we find severe failures on specific subcategories across eight reward models. To mitigate these failures, we propose HARVE, a training-free reward-head editing method for scalar reward models. Instead of fine-tuning the reward model, HARVE identifies a multi-directional hacking subspace from residual stream direction"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.03131","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/2606.03131/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":"2606.03131","created_at":"2026-06-03T01:05:32.537755+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.03131v1","created_at":"2026-06-03T01:05:32.537755+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.03131","created_at":"2026-06-03T01:05:32.537755+00:00"},{"alias_kind":"pith_short_12","alias_value":"WWJFZ7B4WPXR","created_at":"2026-06-03T01:05:32.537755+00:00"},{"alias_kind":"pith_short_16","alias_value":"WWJFZ7B4WPXRHL3T","created_at":"2026-06-03T01:05:32.537755+00:00"},{"alias_kind":"pith_short_8","alias_value":"WWJFZ7B4","created_at":"2026-06-03T01:05:32.537755+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/WWJFZ7B4WPXRHL3TYTN55IPQVI","json":"https://pith.science/pith/WWJFZ7B4WPXRHL3TYTN55IPQVI.json","graph_json":"https://pith.science/api/pith-number/WWJFZ7B4WPXRHL3TYTN55IPQVI/graph.json","events_json":"https://pith.science/api/pith-number/WWJFZ7B4WPXRHL3TYTN55IPQVI/events.json","paper":"https://pith.science/paper/WWJFZ7B4"},"agent_actions":{"view_html":"https://pith.science/pith/WWJFZ7B4WPXRHL3TYTN55IPQVI","download_json":"https://pith.science/pith/WWJFZ7B4WPXRHL3TYTN55IPQVI.json","view_paper":"https://pith.science/paper/WWJFZ7B4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.03131&json=true","fetch_graph":"https://pith.science/api/pith-number/WWJFZ7B4WPXRHL3TYTN55IPQVI/graph.json","fetch_events":"https://pith.science/api/pith-number/WWJFZ7B4WPXRHL3TYTN55IPQVI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WWJFZ7B4WPXRHL3TYTN55IPQVI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WWJFZ7B4WPXRHL3TYTN55IPQVI/action/storage_attestation","attest_author":"https://pith.science/pith/WWJFZ7B4WPXRHL3TYTN55IPQVI/action/author_attestation","sign_citation":"https://pith.science/pith/WWJFZ7B4WPXRHL3TYTN55IPQVI/action/citation_signature","submit_replication":"https://pith.science/pith/WWJFZ7B4WPXRHL3TYTN55IPQVI/action/replication_record"}},"created_at":"2026-06-03T01:05:32.537755+00:00","updated_at":"2026-06-03T01:05:32.537755+00:00"}