{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GTBKDWMZMVDJ2VH6VVS5OZHGRM","short_pith_number":"pith:GTBKDWMZ","schema_version":"1.0","canonical_sha256":"34c2a1d99965469d54fead65d764e68b0dc6f2ca31fe1689c4bceac8956c0324","source":{"kind":"arxiv","id":"2503.12188","version":2},"attestation_state":"computed","paper":{"title":"Multi-Agent Systems Execute Arbitrary Malicious Code","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Harold Triedman, Rishi Jha, Vitaly Shmatikov","submitted_at":"2025-03-15T16:16:08Z","abstract_excerpt":"Multi-agent systems coordinate LLM-based agents to perform tasks on users' behalf. In real-world applications, multi-agent systems will inevitably interact with untrusted inputs, such as malicious Web content, files, email attachments, and more.\n  Using several recently proposed multi-agent frameworks as concrete examples, we demonstrate that adversarial content can hijack control and communication within the system to invoke unsafe agents and functionalities. This results in a complete security breach, up to execution of arbitrary malicious code on the user's device or exfiltration of sensiti"},"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":"2503.12188","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CR","submitted_at":"2025-03-15T16:16:08Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"0b2386b69a74b060d890e8801abac1f6470f127367395c1fafadeae30522990e","abstract_canon_sha256":"ef63d60547f9a585af30dd0607d7793f5a42e06b079108a80f6e71e402c24db3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:11:16.071146Z","signature_b64":"eXIhVJACZ3qfOyCyXmOtvVhn4HAhXqzU6U7yTdYLoQXJlhih5L3L3UEYbRS6wfgO8/bE19sKx47h2p0AG10+Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"34c2a1d99965469d54fead65d764e68b0dc6f2ca31fe1689c4bceac8956c0324","last_reissued_at":"2026-07-05T12:11:16.070585Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:11:16.070585Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Agent Systems Execute Arbitrary Malicious Code","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Harold Triedman, Rishi Jha, Vitaly Shmatikov","submitted_at":"2025-03-15T16:16:08Z","abstract_excerpt":"Multi-agent systems coordinate LLM-based agents to perform tasks on users' behalf. In real-world applications, multi-agent systems will inevitably interact with untrusted inputs, such as malicious Web content, files, email attachments, and more.\n  Using several recently proposed multi-agent frameworks as concrete examples, we demonstrate that adversarial content can hijack control and communication within the system to invoke unsafe agents and functionalities. This results in a complete security breach, up to execution of arbitrary malicious code on the user's device or exfiltration of sensiti"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.12188","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/2503.12188/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":"2503.12188","created_at":"2026-07-05T12:11:16.070655+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.12188v2","created_at":"2026-07-05T12:11:16.070655+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.12188","created_at":"2026-07-05T12:11:16.070655+00:00"},{"alias_kind":"pith_short_12","alias_value":"GTBKDWMZMVDJ","created_at":"2026-07-05T12:11:16.070655+00:00"},{"alias_kind":"pith_short_16","alias_value":"GTBKDWMZMVDJ2VH6","created_at":"2026-07-05T12:11:16.070655+00:00"},{"alias_kind":"pith_short_8","alias_value":"GTBKDWMZ","created_at":"2026-07-05T12:11:16.070655+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07097","citing_title":"Operational Reframing and Approval-Framed Delegation in Multi-Agent LLM Safety","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2606.12797","citing_title":"The Containment Gap: How Deployed Agentic AI Frameworks Fail Public-Facing Safety Requirements","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30602","citing_title":"MESA: Prioritizing Vulnerable Communication Channels for Securing Multi-Agent Systems","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00497","citing_title":"\"I Strongly Suspect This Website Is a Scam\": Benchmarking PII Leakage and Detection without Defense in Autonomous Web Agents","ref_index":151,"is_internal_anchor":false},{"citing_arxiv_id":"2606.10749","citing_title":"Toward Secure LLM Agents: Threat Surfaces, Attacks, Defenses, and Evaluation","ref_index":169,"is_internal_anchor":false},{"citing_arxiv_id":"2601.03294","citing_title":"AgentMark: Utility-Preserving Behavioral Watermarking for Agents","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2603.04474","citing_title":"From Spark to Fire: Modeling and Mitigating Error Cascades in LLM-Based Multi-Agent Collaboration","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2603.09002","citing_title":"Security Considerations for Multi-agent Systems","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11036","citing_title":"Sequential Behavioral Watermarking for LLM Agents","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08460","citing_title":"When Child Inherits: Modeling and Exploiting Subagent Spawn in Multi-Agent Networks","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03378","citing_title":"ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection","ref_index":143,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08608","citing_title":"Semantic Intent Fragmentation: A Single-Shot Compositional Attack on Multi-Agent AI Pipelines","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14317","citing_title":"Challenges and Future Directions in Agentic Reverse Engineering Systems","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16543","citing_title":"Conjunctive Prompt Attacks in Multi-Agent LLM Systems","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM","json":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM.json","graph_json":"https://pith.science/api/pith-number/GTBKDWMZMVDJ2VH6VVS5OZHGRM/graph.json","events_json":"https://pith.science/api/pith-number/GTBKDWMZMVDJ2VH6VVS5OZHGRM/events.json","paper":"https://pith.science/paper/GTBKDWMZ"},"agent_actions":{"view_html":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM","download_json":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM.json","view_paper":"https://pith.science/paper/GTBKDWMZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.12188&json=true","fetch_graph":"https://pith.science/api/pith-number/GTBKDWMZMVDJ2VH6VVS5OZHGRM/graph.json","fetch_events":"https://pith.science/api/pith-number/GTBKDWMZMVDJ2VH6VVS5OZHGRM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM/action/storage_attestation","attest_author":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM/action/author_attestation","sign_citation":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM/action/citation_signature","submit_replication":"https://pith.science/pith/GTBKDWMZMVDJ2VH6VVS5OZHGRM/action/replication_record"}},"created_at":"2026-07-05T12:11:16.070655+00:00","updated_at":"2026-07-05T12:11:16.070655+00:00"}