{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:6GCVRNP2GAVO4RZCC6SZQ4MZC5","short_pith_number":"pith:6GCVRNP2","schema_version":"1.0","canonical_sha256":"f18558b5fa302aee472217a598719917522540950a3ddc0fb5b53d514b406c9b","source":{"kind":"arxiv","id":"2606.15057","version":2},"attestation_state":"computed","paper":{"title":"AutoDojo: Adaptive Black-Box Attacks Reveal the Limits of IPI Defenses and Task-Specification Effects in LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Chaowei Xiao, Ning Zhang, Taoran Li, Xinhang Ma, Yevgeniy Vorobeychik, Zhiyuan Yu","submitted_at":"2026-06-13T02:09:08Z","abstract_excerpt":"Indirect prompt injection (IPI) is a major security threat to LLM-powered agents. Thus, a growing body of work have proposed a variety of defensive approaches against IPI. These can be grouped into three broad categories: 1) prompt-based (using prompting as a way to prevent agents from following malicious instructions), 2) detection-based (identifying and filtering malicious instructions), and 3) system-level (using systems insights, such as control and data isolation, for defense). However, commonly used benchmarks for evaluating defense, such as AgentDojo, are \\emph{inherently static}, gener"},"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":"2606.15057","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2026-06-13T02:09:08Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cb301cfd2f5662135eceb6ae14be2ed1c3ba7b61d9a8d8ae173bdf84a936cc11","abstract_canon_sha256":"e453c798e2a5cd2b152d76e3be8c49ff127f97a74924466de4bc6326290a79cc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T01:12:09.694475Z","signature_b64":"XU6BnoeN4uf1WPLUGZz9fKiCHw3ftv6NjYY6IchAqPsrkpUsgAhsYqqTKZy6ynFSANVg5wSrYdtU97M7wmJeAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f18558b5fa302aee472217a598719917522540950a3ddc0fb5b53d514b406c9b","last_reissued_at":"2026-06-23T01:12:09.693934Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T01:12:09.693934Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AutoDojo: Adaptive Black-Box Attacks Reveal the Limits of IPI Defenses and Task-Specification Effects in LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Chaowei Xiao, Ning Zhang, Taoran Li, Xinhang Ma, Yevgeniy Vorobeychik, Zhiyuan Yu","submitted_at":"2026-06-13T02:09:08Z","abstract_excerpt":"Indirect prompt injection (IPI) is a major security threat to LLM-powered agents. Thus, a growing body of work have proposed a variety of defensive approaches against IPI. These can be grouped into three broad categories: 1) prompt-based (using prompting as a way to prevent agents from following malicious instructions), 2) detection-based (identifying and filtering malicious instructions), and 3) system-level (using systems insights, such as control and data isolation, for defense). However, commonly used benchmarks for evaluating defense, such as AgentDojo, are \\emph{inherently static}, gener"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.15057","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/2606.15057/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":"2606.15057","created_at":"2026-06-23T01:12:09.694010+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.15057v2","created_at":"2026-06-23T01:12:09.694010+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.15057","created_at":"2026-06-23T01:12:09.694010+00:00"},{"alias_kind":"pith_short_12","alias_value":"6GCVRNP2GAVO","created_at":"2026-06-23T01:12:09.694010+00:00"},{"alias_kind":"pith_short_16","alias_value":"6GCVRNP2GAVO4RZC","created_at":"2026-06-23T01:12:09.694010+00:00"},{"alias_kind":"pith_short_8","alias_value":"6GCVRNP2","created_at":"2026-06-23T01:12:09.694010+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/6GCVRNP2GAVO4RZCC6SZQ4MZC5","json":"https://pith.science/pith/6GCVRNP2GAVO4RZCC6SZQ4MZC5.json","graph_json":"https://pith.science/api/pith-number/6GCVRNP2GAVO4RZCC6SZQ4MZC5/graph.json","events_json":"https://pith.science/api/pith-number/6GCVRNP2GAVO4RZCC6SZQ4MZC5/events.json","paper":"https://pith.science/paper/6GCVRNP2"},"agent_actions":{"view_html":"https://pith.science/pith/6GCVRNP2GAVO4RZCC6SZQ4MZC5","download_json":"https://pith.science/pith/6GCVRNP2GAVO4RZCC6SZQ4MZC5.json","view_paper":"https://pith.science/paper/6GCVRNP2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.15057&json=true","fetch_graph":"https://pith.science/api/pith-number/6GCVRNP2GAVO4RZCC6SZQ4MZC5/graph.json","fetch_events":"https://pith.science/api/pith-number/6GCVRNP2GAVO4RZCC6SZQ4MZC5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6GCVRNP2GAVO4RZCC6SZQ4MZC5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6GCVRNP2GAVO4RZCC6SZQ4MZC5/action/storage_attestation","attest_author":"https://pith.science/pith/6GCVRNP2GAVO4RZCC6SZQ4MZC5/action/author_attestation","sign_citation":"https://pith.science/pith/6GCVRNP2GAVO4RZCC6SZQ4MZC5/action/citation_signature","submit_replication":"https://pith.science/pith/6GCVRNP2GAVO4RZCC6SZQ4MZC5/action/replication_record"}},"created_at":"2026-06-23T01:12:09.694010+00:00","updated_at":"2026-06-23T01:12:09.694010+00:00"}