{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CDXRJWZJW2J2F3IXSFQZPOS3D6","short_pith_number":"pith:CDXRJWZJ","schema_version":"1.0","canonical_sha256":"10ef14db29b693a2ed17916197ba5b1fb6501b0111d455306d4e084da202cceb","source":{"kind":"arxiv","id":"2507.04883","version":1},"attestation_state":"computed","paper":{"title":"Beyond Training-time Poisoning: Component-level and Post-training Backdoors in Deep Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CR"],"primary_cat":"cs.LG","authors_text":"Alberto Caron, Chris Hicks, Pete Burnap, Sanyam Vyas, Vasilios Mavroudis","submitted_at":"2025-07-07T11:15:54Z","abstract_excerpt":"Deep Reinforcement Learning (DRL) systems are increasingly used in safety-critical applications, yet their security remains severely underexplored. This work investigates backdoor attacks, which implant hidden triggers that cause malicious actions only when specific inputs appear in the observation space. Existing DRL backdoor research focuses solely on training-time attacks requiring unrealistic access to the training pipeline. In contrast, we reveal critical vulnerabilities across the DRL supply chain where backdoors can be embedded with significantly reduced adversarial privileges. We intro"},"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":"2507.04883","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-07T11:15:54Z","cross_cats_sorted":["cs.AI","cs.CR"],"title_canon_sha256":"73eee7f810321b6cf1f01dd70c09a9c033d1ee2780998fe37b5845d7a20644a2","abstract_canon_sha256":"63eb73a9d6b1dab594e7cb1bf7aeda134c00c2f5757f468a709439fa39517e68"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:32:55.811980Z","signature_b64":"MgiQ1sWd57MrPOLKhzH865C5ubmUQ6MZ7EpaXH9da+xuhuVpjU88YC5HHxsULOJdTrN5gCGg0+a2Pcs3bDhNCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"10ef14db29b693a2ed17916197ba5b1fb6501b0111d455306d4e084da202cceb","last_reissued_at":"2026-07-05T11:32:55.811468Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:32:55.811468Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond Training-time Poisoning: Component-level and Post-training Backdoors in Deep Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CR"],"primary_cat":"cs.LG","authors_text":"Alberto Caron, Chris Hicks, Pete Burnap, Sanyam Vyas, Vasilios Mavroudis","submitted_at":"2025-07-07T11:15:54Z","abstract_excerpt":"Deep Reinforcement Learning (DRL) systems are increasingly used in safety-critical applications, yet their security remains severely underexplored. This work investigates backdoor attacks, which implant hidden triggers that cause malicious actions only when specific inputs appear in the observation space. Existing DRL backdoor research focuses solely on training-time attacks requiring unrealistic access to the training pipeline. In contrast, we reveal critical vulnerabilities across the DRL supply chain where backdoors can be embedded with significantly reduced adversarial privileges. We intro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.04883","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/2507.04883/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":"2507.04883","created_at":"2026-07-05T11:32:55.811541+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.04883v1","created_at":"2026-07-05T11:32:55.811541+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.04883","created_at":"2026-07-05T11:32:55.811541+00:00"},{"alias_kind":"pith_short_12","alias_value":"CDXRJWZJW2J2","created_at":"2026-07-05T11:32:55.811541+00:00"},{"alias_kind":"pith_short_16","alias_value":"CDXRJWZJW2J2F3IX","created_at":"2026-07-05T11:32:55.811541+00:00"},{"alias_kind":"pith_short_8","alias_value":"CDXRJWZJ","created_at":"2026-07-05T11:32:55.811541+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.09638","citing_title":"Plan2Cleanse: Test-Time Backdoor Defense via Monte-Carlo Planning in Deep Reinforcement Learning","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6","json":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6.json","graph_json":"https://pith.science/api/pith-number/CDXRJWZJW2J2F3IXSFQZPOS3D6/graph.json","events_json":"https://pith.science/api/pith-number/CDXRJWZJW2J2F3IXSFQZPOS3D6/events.json","paper":"https://pith.science/paper/CDXRJWZJ"},"agent_actions":{"view_html":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6","download_json":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6.json","view_paper":"https://pith.science/paper/CDXRJWZJ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.04883&json=true","fetch_graph":"https://pith.science/api/pith-number/CDXRJWZJW2J2F3IXSFQZPOS3D6/graph.json","fetch_events":"https://pith.science/api/pith-number/CDXRJWZJW2J2F3IXSFQZPOS3D6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6/action/storage_attestation","attest_author":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6/action/author_attestation","sign_citation":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6/action/citation_signature","submit_replication":"https://pith.science/pith/CDXRJWZJW2J2F3IXSFQZPOS3D6/action/replication_record"}},"created_at":"2026-07-05T11:32:55.811541+00:00","updated_at":"2026-07-05T11:32:55.811541+00:00"}