{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XXOJ3FF4ECPCR4JKU2OXJUYFLV","short_pith_number":"pith:XXOJ3FF4","schema_version":"1.0","canonical_sha256":"bddc9d94bc209e28f12aa69d74d3055d643e1fd3272dbd724e0016f68a9d7613","source":{"kind":"arxiv","id":"2502.09809","version":1},"attestation_state":"computed","paper":{"title":"AgentGuard: Repurposing Agentic Orchestrator for Safety Evaluation of Tool Orchestration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Jizhou Chen, Samuel Lee Cong","submitted_at":"2025-02-13T23:00:33Z","abstract_excerpt":"The integration of tool use into large language models (LLMs) enables agentic systems with real-world impact. In the meantime, unlike standalone LLMs, compromised agents can execute malicious workflows with more consequential impact, signified by their tool-use capability. We propose AgentGuard, a framework to autonomously discover and validate unsafe tool-use workflows, followed by generating safety constraints to confine the behaviors of agents, achieving the baseline of safety guarantee at deployment. AgentGuard leverages the LLM orchestrator's innate capabilities - knowledge of tool functi"},"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":"2502.09809","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2025-02-13T23:00:33Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"077af22b987011082792eb28f38557f30081d18a117c617e9f3f5fd459110f8d","abstract_canon_sha256":"ef83be5f19cda2df5513308ffb12d9441654e2270e65a86dc2af5ad0f0744aa9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:14:12.465067Z","signature_b64":"8pwMI5TFjGYxuzrjOduWgDMyPwSHiRKw0FYGgefl3WYGePJfjRfkQkGK+B8CkOtmrSvbb3BXZUlHGgv8/UzOCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bddc9d94bc209e28f12aa69d74d3055d643e1fd3272dbd724e0016f68a9d7613","last_reissued_at":"2026-07-05T10:14:12.464592Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:14:12.464592Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AgentGuard: Repurposing Agentic Orchestrator for Safety Evaluation of Tool Orchestration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"Jizhou Chen, Samuel Lee Cong","submitted_at":"2025-02-13T23:00:33Z","abstract_excerpt":"The integration of tool use into large language models (LLMs) enables agentic systems with real-world impact. In the meantime, unlike standalone LLMs, compromised agents can execute malicious workflows with more consequential impact, signified by their tool-use capability. We propose AgentGuard, a framework to autonomously discover and validate unsafe tool-use workflows, followed by generating safety constraints to confine the behaviors of agents, achieving the baseline of safety guarantee at deployment. AgentGuard leverages the LLM orchestrator's innate capabilities - knowledge of tool functi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.09809","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/2502.09809/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":"2502.09809","created_at":"2026-07-05T10:14:12.464649+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.09809v1","created_at":"2026-07-05T10:14:12.464649+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.09809","created_at":"2026-07-05T10:14:12.464649+00:00"},{"alias_kind":"pith_short_12","alias_value":"XXOJ3FF4ECPC","created_at":"2026-07-05T10:14:12.464649+00:00"},{"alias_kind":"pith_short_16","alias_value":"XXOJ3FF4ECPCR4JK","created_at":"2026-07-05T10:14:12.464649+00:00"},{"alias_kind":"pith_short_8","alias_value":"XXOJ3FF4","created_at":"2026-07-05T10:14:12.464649+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10749","citing_title":"Toward Secure LLM Agents: Threat Surfaces, Attacks, Defenses, and Evaluation","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2603.09002","citing_title":"Security Considerations for Multi-agent Systems","ref_index":300,"is_internal_anchor":false},{"citing_arxiv_id":"2503.23278","citing_title":"Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV","json":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV.json","graph_json":"https://pith.science/api/pith-number/XXOJ3FF4ECPCR4JKU2OXJUYFLV/graph.json","events_json":"https://pith.science/api/pith-number/XXOJ3FF4ECPCR4JKU2OXJUYFLV/events.json","paper":"https://pith.science/paper/XXOJ3FF4"},"agent_actions":{"view_html":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV","download_json":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV.json","view_paper":"https://pith.science/paper/XXOJ3FF4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.09809&json=true","fetch_graph":"https://pith.science/api/pith-number/XXOJ3FF4ECPCR4JKU2OXJUYFLV/graph.json","fetch_events":"https://pith.science/api/pith-number/XXOJ3FF4ECPCR4JKU2OXJUYFLV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV/action/storage_attestation","attest_author":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV/action/author_attestation","sign_citation":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV/action/citation_signature","submit_replication":"https://pith.science/pith/XXOJ3FF4ECPCR4JKU2OXJUYFLV/action/replication_record"}},"created_at":"2026-07-05T10:14:12.464649+00:00","updated_at":"2026-07-05T10:14:12.464649+00:00"}