{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:D5APOQ4LJ4MEJYQ2YQCISBLXEY","short_pith_number":"pith:D5APOQ4L","schema_version":"1.0","canonical_sha256":"1f40f7438b4f1844e21ac4048905772601dbab750851569487f266fabd511dfc","source":{"kind":"arxiv","id":"2607.27574","version":1},"attestation_state":"computed","paper":{"title":"Policy Gradient Steering: Interventions from Behavioral Objectives","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.MA"],"primary_cat":"cs.LG","authors_text":"Aur\\'elie Beynier, Nicolas Maudet, Yoann Poupart","submitted_at":"2026-07-30T01:29:57Z","abstract_excerpt":"Activation steering has emerged in large language models as a lightweight alternative for dynamically changing a model's behavior at inference time. However, we show that existing steering methods fail to steer even a simple policy in a two-route gridworld environment. To address this limitation, we propose Policy Gradient Steering (PGS), which formulates steering as a reinforcement learning problem. PGS accumulates gradients of a temporary behavioral objective over a small set of rollouts or demonstrations to construct a removable task vector. We first demonstrate the calibration and reversib"},"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":"2607.27574","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-07-30T01:29:57Z","cross_cats_sorted":["cs.MA"],"title_canon_sha256":"469e01747253b071dbe2889e59aab2bfb475dbe7591e6a3b106c790032897209","abstract_canon_sha256":"5a6401e246c0014263805fb8acd2dca008fa583757bcbd0af4314ff086b005a4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1f40f7438b4f1844e21ac4048905772601dbab750851569487f266fabd511dfc","last_reissued_at":"2026-07-31T01:26:24.961906Z","signature_status":"unsigned_v0","first_computed_at":"2026-07-31T01:26:24.961906Z"},"graph_snapshot":{"paper":{"title":"Policy Gradient Steering: Interventions from Behavioral Objectives","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.MA"],"primary_cat":"cs.LG","authors_text":"Aur\\'elie Beynier, Nicolas Maudet, Yoann Poupart","submitted_at":"2026-07-30T01:29:57Z","abstract_excerpt":"Activation steering has emerged in large language models as a lightweight alternative for dynamically changing a model's behavior at inference time. However, we show that existing steering methods fail to steer even a simple policy in a two-route gridworld environment. To address this limitation, we propose Policy Gradient Steering (PGS), which formulates steering as a reinforcement learning problem. PGS accumulates gradients of a temporary behavioral objective over a small set of rollouts or demonstrations to construct a removable task vector. We first demonstrate the calibration and reversib"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.27574","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/2607.27574/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":"2607.27574","created_at":"2026-07-31T01:26:24.965003+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.27574v1","created_at":"2026-07-31T01:26:24.965003+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.27574","created_at":"2026-07-31T01:26:24.965003+00:00"},{"alias_kind":"pith_short_12","alias_value":"D5APOQ4LJ4ME","created_at":"2026-07-31T01:26:24.965003+00:00"},{"alias_kind":"pith_short_16","alias_value":"D5APOQ4LJ4MEJYQ2","created_at":"2026-07-31T01:26:24.965003+00:00"},{"alias_kind":"pith_short_8","alias_value":"D5APOQ4L","created_at":"2026-07-31T01:26:24.965003+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/D5APOQ4LJ4MEJYQ2YQCISBLXEY","json":"https://pith.science/pith/D5APOQ4LJ4MEJYQ2YQCISBLXEY.json","graph_json":"https://pith.science/api/pith-number/D5APOQ4LJ4MEJYQ2YQCISBLXEY/graph.json","events_json":"https://pith.science/api/pith-number/D5APOQ4LJ4MEJYQ2YQCISBLXEY/events.json","paper":"https://pith.science/paper/D5APOQ4L"},"agent_actions":{"view_html":"https://pith.science/pith/D5APOQ4LJ4MEJYQ2YQCISBLXEY","download_json":"https://pith.science/pith/D5APOQ4LJ4MEJYQ2YQCISBLXEY.json","view_paper":"https://pith.science/paper/D5APOQ4L","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.27574&json=true","fetch_graph":"https://pith.science/api/pith-number/D5APOQ4LJ4MEJYQ2YQCISBLXEY/graph.json","fetch_events":"https://pith.science/api/pith-number/D5APOQ4LJ4MEJYQ2YQCISBLXEY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/D5APOQ4LJ4MEJYQ2YQCISBLXEY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/D5APOQ4LJ4MEJYQ2YQCISBLXEY/action/storage_attestation","attest_author":"https://pith.science/pith/D5APOQ4LJ4MEJYQ2YQCISBLXEY/action/author_attestation","sign_citation":"https://pith.science/pith/D5APOQ4LJ4MEJYQ2YQCISBLXEY/action/citation_signature","submit_replication":"https://pith.science/pith/D5APOQ4LJ4MEJYQ2YQCISBLXEY/action/replication_record"}},"created_at":"2026-07-31T01:26:24.965003+00:00","updated_at":"2026-07-31T01:26:24.965003+00:00"}