{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:INUB7BRGWKOEO4CKHHYJVVEYTT","short_pith_number":"pith:INUB7BRG","schema_version":"1.0","canonical_sha256":"43681f8626b29c47704a39f09ad4989cd13a7ad6a99800822771c155339bb4c9","source":{"kind":"arxiv","id":"2505.19532","version":2},"attestation_state":"computed","paper":{"title":"Fox in the Henhouse: Supply-Chain Backdoor Attacks Against Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andrew C. Cullen, Benjamin I. P. Rubinstein, Paul Montague, Sarah Erfani, Shijie Liu","submitted_at":"2025-05-26T05:39:35Z","abstract_excerpt":"The current state-of-the-art backdoor attacks against Reinforcement Learning (RL) rely upon unrealistically permissive access models, that assume the attacker can read (or even write) the victim's policy parameters, observations, or rewards. In this work, we question whether such a strong assumption is required to launch backdoor attacks against RL. To answer this question, we propose the \\underline{S}upply-\\underline{C}h\\underline{a}in \\underline{B}ackdoor (SCAB) attack, which targets a common RL workflow: training agents using external agents that are provided separately or embedded within t"},"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":"2505.19532","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-26T05:39:35Z","cross_cats_sorted":[],"title_canon_sha256":"4ddb90026c0113a674f2fda7a5e6eefd5fe1dbbdf23768d69a44d2f25b34c013","abstract_canon_sha256":"ba85e52d6295d868ac84b10325a6358f5387dd2684a5025e9de631cda52faf7f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-25T01:17:44.712821Z","signature_b64":"hZZLGhCKgINKq2VT8LE2xNaiqbr5Yw0VCwdb5bk7zVHXXzFsp4H2Mp8+IeZVysoan7ZpovNqHG8lxNdqpTjXDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"43681f8626b29c47704a39f09ad4989cd13a7ad6a99800822771c155339bb4c9","last_reissued_at":"2026-06-25T01:17:44.712337Z","signature_status":"signed_v1","first_computed_at":"2026-06-25T01:17:44.712337Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fox in the Henhouse: Supply-Chain Backdoor Attacks Against Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Andrew C. Cullen, Benjamin I. P. Rubinstein, Paul Montague, Sarah Erfani, Shijie Liu","submitted_at":"2025-05-26T05:39:35Z","abstract_excerpt":"The current state-of-the-art backdoor attacks against Reinforcement Learning (RL) rely upon unrealistically permissive access models, that assume the attacker can read (or even write) the victim's policy parameters, observations, or rewards. In this work, we question whether such a strong assumption is required to launch backdoor attacks against RL. To answer this question, we propose the \\underline{S}upply-\\underline{C}h\\underline{a}in \\underline{B}ackdoor (SCAB) attack, which targets a common RL workflow: training agents using external agents that are provided separately or embedded within t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.19532","kind":"arxiv","version":2},"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/2505.19532/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":"2505.19532","created_at":"2026-06-25T01:17:44.712396+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.19532v2","created_at":"2026-06-25T01:17:44.712396+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.19532","created_at":"2026-06-25T01:17:44.712396+00:00"},{"alias_kind":"pith_short_12","alias_value":"INUB7BRGWKOE","created_at":"2026-06-25T01:17:44.712396+00:00"},{"alias_kind":"pith_short_16","alias_value":"INUB7BRGWKOEO4CK","created_at":"2026-06-25T01:17:44.712396+00:00"},{"alias_kind":"pith_short_8","alias_value":"INUB7BRG","created_at":"2026-06-25T01:17:44.712396+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.09638","citing_title":"Plan2Cleanse: Test-Time Backdoor Defense via Monte-Carlo Planning in Deep Reinforcement Learning","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2605.05977","citing_title":"BehaviorGuard: Online Backdoor Defense for Deep Reinforcement Learning","ref_index":20,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT","json":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT.json","graph_json":"https://pith.science/api/pith-number/INUB7BRGWKOEO4CKHHYJVVEYTT/graph.json","events_json":"https://pith.science/api/pith-number/INUB7BRGWKOEO4CKHHYJVVEYTT/events.json","paper":"https://pith.science/paper/INUB7BRG"},"agent_actions":{"view_html":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT","download_json":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT.json","view_paper":"https://pith.science/paper/INUB7BRG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.19532&json=true","fetch_graph":"https://pith.science/api/pith-number/INUB7BRGWKOEO4CKHHYJVVEYTT/graph.json","fetch_events":"https://pith.science/api/pith-number/INUB7BRGWKOEO4CKHHYJVVEYTT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT/action/storage_attestation","attest_author":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT/action/author_attestation","sign_citation":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT/action/citation_signature","submit_replication":"https://pith.science/pith/INUB7BRGWKOEO4CKHHYJVVEYTT/action/replication_record"}},"created_at":"2026-06-25T01:17:44.712396+00:00","updated_at":"2026-06-25T01:17:44.712396+00:00"}