{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:E3LC6YJVUXMILZAMIYRRDMVHP6","short_pith_number":"pith:E3LC6YJV","schema_version":"1.0","canonical_sha256":"26d62f6135a5d885e40c462311b2a77f8166ed497faafb679bcb3ee8d088fb77","source":{"kind":"arxiv","id":"2606.06223","version":1},"attestation_state":"computed","paper":{"title":"From Reward-Hack Activations to Agentic Risk States: Context-Calibrated Mechanistic Monitoring in LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Odej Kao, Patrick Wilhelm","submitted_at":"2026-06-04T14:34:31Z","abstract_excerpt":"Language-model agents act through repeated cycles of observation, reasoning, and action selection, making safety monitoring depend on both internal model state and environment context. We study reward-hacking monitors in ReAct-style agents acting in Gameable ALFWorld and WebShop. Agents are instrumented with activation-based reward-hack scores, token-level entropy, and decision-context features. We find that adapters fine-tuned on \\textit{School-of-Reward-Hacks} dataset can transfer reward-hack tendencies into agentic action selection, especially when the environment exposes proxy-reward affor"},"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":"2606.06223","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-06-04T14:34:31Z","cross_cats_sorted":[],"title_canon_sha256":"d21a33b9bf284f1cd0fdb36ae2e294075f46c4c22aa5854a598806fadd17af76","abstract_canon_sha256":"96054256ae0c742b9f9aa04020ed5e7d4c5f8e70b9ece8ff11273ec5a8faa57a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:15:38.316086Z","signature_b64":"S8bkoSNF4GxA6NS6UZRTjQY0QxKGMR//zqYQNo/liomV0b0YWvAEL4sn8wAXBNCzvt8eoqOvzMQGC5QHSDpGDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"26d62f6135a5d885e40c462311b2a77f8166ed497faafb679bcb3ee8d088fb77","last_reissued_at":"2026-06-05T01:15:38.315426Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:15:38.315426Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From Reward-Hack Activations to Agentic Risk States: Context-Calibrated Mechanistic Monitoring in LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Odej Kao, Patrick Wilhelm","submitted_at":"2026-06-04T14:34:31Z","abstract_excerpt":"Language-model agents act through repeated cycles of observation, reasoning, and action selection, making safety monitoring depend on both internal model state and environment context. We study reward-hacking monitors in ReAct-style agents acting in Gameable ALFWorld and WebShop. Agents are instrumented with activation-based reward-hack scores, token-level entropy, and decision-context features. We find that adapters fine-tuned on \\textit{School-of-Reward-Hacks} dataset can transfer reward-hack tendencies into agentic action selection, especially when the environment exposes proxy-reward affor"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.06223","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/2606.06223/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":"2606.06223","created_at":"2026-06-05T01:15:38.315535+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.06223v1","created_at":"2026-06-05T01:15:38.315535+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.06223","created_at":"2026-06-05T01:15:38.315535+00:00"},{"alias_kind":"pith_short_12","alias_value":"E3LC6YJVUXMI","created_at":"2026-06-05T01:15:38.315535+00:00"},{"alias_kind":"pith_short_16","alias_value":"E3LC6YJVUXMILZAM","created_at":"2026-06-05T01:15:38.315535+00:00"},{"alias_kind":"pith_short_8","alias_value":"E3LC6YJV","created_at":"2026-06-05T01:15:38.315535+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/E3LC6YJVUXMILZAMIYRRDMVHP6","json":"https://pith.science/pith/E3LC6YJVUXMILZAMIYRRDMVHP6.json","graph_json":"https://pith.science/api/pith-number/E3LC6YJVUXMILZAMIYRRDMVHP6/graph.json","events_json":"https://pith.science/api/pith-number/E3LC6YJVUXMILZAMIYRRDMVHP6/events.json","paper":"https://pith.science/paper/E3LC6YJV"},"agent_actions":{"view_html":"https://pith.science/pith/E3LC6YJVUXMILZAMIYRRDMVHP6","download_json":"https://pith.science/pith/E3LC6YJVUXMILZAMIYRRDMVHP6.json","view_paper":"https://pith.science/paper/E3LC6YJV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.06223&json=true","fetch_graph":"https://pith.science/api/pith-number/E3LC6YJVUXMILZAMIYRRDMVHP6/graph.json","fetch_events":"https://pith.science/api/pith-number/E3LC6YJVUXMILZAMIYRRDMVHP6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/E3LC6YJVUXMILZAMIYRRDMVHP6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/E3LC6YJVUXMILZAMIYRRDMVHP6/action/storage_attestation","attest_author":"https://pith.science/pith/E3LC6YJVUXMILZAMIYRRDMVHP6/action/author_attestation","sign_citation":"https://pith.science/pith/E3LC6YJVUXMILZAMIYRRDMVHP6/action/citation_signature","submit_replication":"https://pith.science/pith/E3LC6YJVUXMILZAMIYRRDMVHP6/action/replication_record"}},"created_at":"2026-06-05T01:15:38.315535+00:00","updated_at":"2026-06-05T01:15:38.315535+00:00"}