{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:B7JDALTDJPMI5U7W6HEKZPTDQB","short_pith_number":"pith:B7JDALTD","schema_version":"1.0","canonical_sha256":"0fd2302e634bd88ed3f6f1c8acbe6380577e44010410f08e8804bd7b6467a316","source":{"kind":"arxiv","id":"2506.08336","version":2},"attestation_state":"computed","paper":{"title":"Your Agent Can Defend Itself against Backdoor Attacks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CR","authors_text":"Cao Bochuan, Chen Jinghui, Liang Jiacheng, Li Changjiang, Wang Ting","submitted_at":"2025-06-10T01:45:56Z","abstract_excerpt":"Despite their growing adoption across domains, large language model (LLM)-powered agents face significant security risks from backdoor attacks during training and fine-tuning. These compromised agents can subsequently be manipulated to execute malicious operations when presented with specific triggers in their inputs or environments. To address this pressing risk, we present ReAgent, a novel defense against a range of backdoor attacks on LLM-based agents. Intuitively, backdoor attacks often result in inconsistencies among the user's instruction, the agent's planning, and its execution. Drawing"},"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":"2506.08336","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-06-10T01:45:56Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"8bd8ed3e8509262f62ab98e3f5bb23f3ce6d534c047c81208061a919a96492cb","abstract_canon_sha256":"419a92d0e1177e3496bf2e236fd0a4801a1062b32ae19997ec938038fd103313"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:29.377923Z","signature_b64":"ZsLM5msWUkqfWnmv9BfXXI03Rwnon8vMVrpK2fC5RfyP0EwtPehoDxpzY25WwRWpFat9kLxM/U1/tWkHWwDEBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0fd2302e634bd88ed3f6f1c8acbe6380577e44010410f08e8804bd7b6467a316","last_reissued_at":"2026-07-05T11:19:29.377418Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:29.377418Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Your Agent Can Defend Itself against Backdoor Attacks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CR","authors_text":"Cao Bochuan, Chen Jinghui, Liang Jiacheng, Li Changjiang, Wang Ting","submitted_at":"2025-06-10T01:45:56Z","abstract_excerpt":"Despite their growing adoption across domains, large language model (LLM)-powered agents face significant security risks from backdoor attacks during training and fine-tuning. These compromised agents can subsequently be manipulated to execute malicious operations when presented with specific triggers in their inputs or environments. To address this pressing risk, we present ReAgent, a novel defense against a range of backdoor attacks on LLM-based agents. Intuitively, backdoor attacks often result in inconsistencies among the user's instruction, the agent's planning, and its execution. Drawing"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.08336","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/2506.08336/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":"2506.08336","created_at":"2026-07-05T11:19:29.377483+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.08336v2","created_at":"2026-07-05T11:19:29.377483+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.08336","created_at":"2026-07-05T11:19:29.377483+00:00"},{"alias_kind":"pith_short_12","alias_value":"B7JDALTDJPMI","created_at":"2026-07-05T11:19:29.377483+00:00"},{"alias_kind":"pith_short_16","alias_value":"B7JDALTDJPMI5U7W","created_at":"2026-07-05T11:19:29.377483+00:00"},{"alias_kind":"pith_short_8","alias_value":"B7JDALTD","created_at":"2026-07-05T11:19:29.377483+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.03890","citing_title":"From Prompt to Physical Action: Structured Backdoor Attacks on LLM-Mediated Robotic Control Systems","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08612","citing_title":"ATAAT: Adaptive Threat-Aware Adversarial Tuning Framework against Backdoor Attacks on Vision-Language-Action Models","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB","json":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB.json","graph_json":"https://pith.science/api/pith-number/B7JDALTDJPMI5U7W6HEKZPTDQB/graph.json","events_json":"https://pith.science/api/pith-number/B7JDALTDJPMI5U7W6HEKZPTDQB/events.json","paper":"https://pith.science/paper/B7JDALTD"},"agent_actions":{"view_html":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB","download_json":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB.json","view_paper":"https://pith.science/paper/B7JDALTD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.08336&json=true","fetch_graph":"https://pith.science/api/pith-number/B7JDALTDJPMI5U7W6HEKZPTDQB/graph.json","fetch_events":"https://pith.science/api/pith-number/B7JDALTDJPMI5U7W6HEKZPTDQB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB/action/storage_attestation","attest_author":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB/action/author_attestation","sign_citation":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB/action/citation_signature","submit_replication":"https://pith.science/pith/B7JDALTDJPMI5U7W6HEKZPTDQB/action/replication_record"}},"created_at":"2026-07-05T11:19:29.377483+00:00","updated_at":"2026-07-05T11:19:29.377483+00:00"}