{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5MUJFDWOGCXCGV66AYJTVQ3TRV","short_pith_number":"pith:5MUJFDWO","schema_version":"1.0","canonical_sha256":"eb28928ece30ae2357de06133ac3738d63382607836f08040a796d6e6738359d","source":{"kind":"arxiv","id":"2506.06891","version":3},"attestation_state":"computed","paper":{"title":"Robust In-Context Reinforcement Learning Under Reward Poisoning Attacks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Goran Radanovi\\'c, Paulius Sasnauskas, Yi\\u{g}it Yal{\\i}n","submitted_at":"2025-06-07T18:39:47Z","abstract_excerpt":"We study the corruption-robustness of in-context reinforcement learning (ICRL), focusing on the Decision-Pretrained Transformer (DPT, Lee et al., 2023). To address the challenge of reward poisoning attacks targeting the DPT, we propose a novel adversarial training framework, called Adversarially Trained DPT (AT-DPT). Our method simultaneously trains a population of attackers to minimize the true reward of the DPT by poisoning environment rewards, and a DPT model to infer optimal actions from the poisoned data. We evaluate the effectiveness of our approach against standard bandit algorithms, in"},"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":"2506.06891","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-06-07T18:39:47Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"957a096dc7a7dd3473bff1184f14e5168d377b84d0bd49af84dd332bac06401b","abstract_canon_sha256":"41377035f0aebcb320ed238cb7ae919e539b414429396d2e12588ed5ed1ae1e3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-09T01:04:38.773155Z","signature_b64":"MOiiTvF+dKHPoZoM5GTOpU0eBojSh8ANzzvubiBxX4WDgKaBEvkM4OJggqoFKBC7FK7aRNfEi15wtQPYKxDcDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"eb28928ece30ae2357de06133ac3738d63382607836f08040a796d6e6738359d","last_reissued_at":"2026-06-09T01:04:38.772569Z","signature_status":"signed_v1","first_computed_at":"2026-06-09T01:04:38.772569Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Robust In-Context Reinforcement Learning Under Reward Poisoning Attacks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.LG","authors_text":"Goran Radanovi\\'c, Paulius Sasnauskas, Yi\\u{g}it Yal{\\i}n","submitted_at":"2025-06-07T18:39:47Z","abstract_excerpt":"We study the corruption-robustness of in-context reinforcement learning (ICRL), focusing on the Decision-Pretrained Transformer (DPT, Lee et al., 2023). To address the challenge of reward poisoning attacks targeting the DPT, we propose a novel adversarial training framework, called Adversarially Trained DPT (AT-DPT). Our method simultaneously trains a population of attackers to minimize the true reward of the DPT by poisoning environment rewards, and a DPT model to infer optimal actions from the poisoned data. We evaluate the effectiveness of our approach against standard bandit algorithms, in"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.06891","kind":"arxiv","version":3},"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/2506.06891/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":"2506.06891","created_at":"2026-06-09T01:04:38.772630+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.06891v3","created_at":"2026-06-09T01:04:38.772630+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.06891","created_at":"2026-06-09T01:04:38.772630+00:00"},{"alias_kind":"pith_short_12","alias_value":"5MUJFDWOGCXC","created_at":"2026-06-09T01:04:38.772630+00:00"},{"alias_kind":"pith_short_16","alias_value":"5MUJFDWOGCXCGV66","created_at":"2026-06-09T01:04:38.772630+00:00"},{"alias_kind":"pith_short_8","alias_value":"5MUJFDWO","created_at":"2026-06-09T01:04:38.772630+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/5MUJFDWOGCXCGV66AYJTVQ3TRV","json":"https://pith.science/pith/5MUJFDWOGCXCGV66AYJTVQ3TRV.json","graph_json":"https://pith.science/api/pith-number/5MUJFDWOGCXCGV66AYJTVQ3TRV/graph.json","events_json":"https://pith.science/api/pith-number/5MUJFDWOGCXCGV66AYJTVQ3TRV/events.json","paper":"https://pith.science/paper/5MUJFDWO"},"agent_actions":{"view_html":"https://pith.science/pith/5MUJFDWOGCXCGV66AYJTVQ3TRV","download_json":"https://pith.science/pith/5MUJFDWOGCXCGV66AYJTVQ3TRV.json","view_paper":"https://pith.science/paper/5MUJFDWO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.06891&json=true","fetch_graph":"https://pith.science/api/pith-number/5MUJFDWOGCXCGV66AYJTVQ3TRV/graph.json","fetch_events":"https://pith.science/api/pith-number/5MUJFDWOGCXCGV66AYJTVQ3TRV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5MUJFDWOGCXCGV66AYJTVQ3TRV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5MUJFDWOGCXCGV66AYJTVQ3TRV/action/storage_attestation","attest_author":"https://pith.science/pith/5MUJFDWOGCXCGV66AYJTVQ3TRV/action/author_attestation","sign_citation":"https://pith.science/pith/5MUJFDWOGCXCGV66AYJTVQ3TRV/action/citation_signature","submit_replication":"https://pith.science/pith/5MUJFDWOGCXCGV66AYJTVQ3TRV/action/replication_record"}},"created_at":"2026-06-09T01:04:38.772630+00:00","updated_at":"2026-06-09T01:04:38.772630+00:00"}