{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:IW7J2RLIX4H7TBMRKLVS44MVNL","short_pith_number":"pith:IW7J2RLI","schema_version":"1.0","canonical_sha256":"45be9d4568bf0ff9859152eb2e71956af0e12ac7a213e8293b31b906acf9425c","source":{"kind":"arxiv","id":"2503.00061","version":2},"attestation_state":"computed","paper":{"title":"Adaptive Attacks Break Defenses Against Indirect Prompt Injection Attacks on LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Daniel Kang, Henil Shalin Panchal, Qiusi Zhan, Richard Fang","submitted_at":"2025-02-27T04:04:50Z","abstract_excerpt":"Large Language Model (LLM) agents exhibit remarkable performance across diverse applications by using external tools to interact with environments. However, integrating external tools introduces security risks, such as indirect prompt injection (IPI) attacks. Despite defenses designed for IPI attacks, their robustness remains questionable due to insufficient testing against adaptive attacks. In this paper, we evaluate eight different defenses and bypass all of them using adaptive attacks, consistently achieving an attack success rate of over 50%. This reveals critical vulnerabilities in curren"},"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":"2503.00061","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-02-27T04:04:50Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"05a66c80de22e14475e1e9a710f58730e6d4bafff3a0c6bd91b5f553bc8e2d25","abstract_canon_sha256":"acc331f66e7c14f673537482511e4c86417dd92c8aedc27e34d3be96cca6a162"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:23:30.396104Z","signature_b64":"cjoerdFmYOe7Kc6q4soaebkvBDhb37o6R9Nh3qIZj+VTJmkjznfSr5q1OJHf3UlgX22Bt23LH85pp93kXmcWCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"45be9d4568bf0ff9859152eb2e71956af0e12ac7a213e8293b31b906acf9425c","last_reissued_at":"2026-07-05T10:23:30.395407Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:23:30.395407Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Adaptive Attacks Break Defenses Against Indirect Prompt Injection Attacks on LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Daniel Kang, Henil Shalin Panchal, Qiusi Zhan, Richard Fang","submitted_at":"2025-02-27T04:04:50Z","abstract_excerpt":"Large Language Model (LLM) agents exhibit remarkable performance across diverse applications by using external tools to interact with environments. However, integrating external tools introduces security risks, such as indirect prompt injection (IPI) attacks. Despite defenses designed for IPI attacks, their robustness remains questionable due to insufficient testing against adaptive attacks. In this paper, we evaluate eight different defenses and bypass all of them using adaptive attacks, consistently achieving an attack success rate of over 50%. This reveals critical vulnerabilities in curren"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.00061","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/2503.00061/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":"2503.00061","created_at":"2026-07-05T10:23:30.395480+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.00061v2","created_at":"2026-07-05T10:23:30.395480+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.00061","created_at":"2026-07-05T10:23:30.395480+00:00"},{"alias_kind":"pith_short_12","alias_value":"IW7J2RLIX4H7","created_at":"2026-07-05T10:23:30.395480+00:00"},{"alias_kind":"pith_short_16","alias_value":"IW7J2RLIX4H7TBMR","created_at":"2026-07-05T10:23:30.395480+00:00"},{"alias_kind":"pith_short_8","alias_value":"IW7J2RLI","created_at":"2026-07-05T10:23:30.395480+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23277","citing_title":"GIF: Locally Sound Geometric Information Flow Control for LLMs","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30783","citing_title":"Security--Fidelity Tradeoffs: The Hidden Cost of Prompt Injection Defense","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00914","citing_title":"Adversarial Feeds Steer LLM Agent Decisions Against Their Defaults","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18248","citing_title":"Beyond Pattern Matching: Seven Cross-Domain Techniques for Prompt Injection Detection","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2603.12277","citing_title":"Prompt Injection as Role Confusion","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18248","citing_title":"Beyond Pattern Matching: Seven Cross-Domain Techniques for Prompt Injection Detection","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL","json":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL.json","graph_json":"https://pith.science/api/pith-number/IW7J2RLIX4H7TBMRKLVS44MVNL/graph.json","events_json":"https://pith.science/api/pith-number/IW7J2RLIX4H7TBMRKLVS44MVNL/events.json","paper":"https://pith.science/paper/IW7J2RLI"},"agent_actions":{"view_html":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL","download_json":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL.json","view_paper":"https://pith.science/paper/IW7J2RLI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.00061&json=true","fetch_graph":"https://pith.science/api/pith-number/IW7J2RLIX4H7TBMRKLVS44MVNL/graph.json","fetch_events":"https://pith.science/api/pith-number/IW7J2RLIX4H7TBMRKLVS44MVNL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL/action/storage_attestation","attest_author":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL/action/author_attestation","sign_citation":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL/action/citation_signature","submit_replication":"https://pith.science/pith/IW7J2RLIX4H7TBMRKLVS44MVNL/action/replication_record"}},"created_at":"2026-07-05T10:23:30.395480+00:00","updated_at":"2026-07-05T10:23:30.395480+00:00"}