{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:B5WK5MVCUGGNLZDHUTF6Q7RBUW","short_pith_number":"pith:B5WK5MVC","schema_version":"1.0","canonical_sha256":"0f6caeb2a2a18cd5e467a4cbe87e21a595bdd798b87fc43750dfed097ac01b2b","source":{"kind":"arxiv","id":"2607.07368","version":1},"attestation_state":"computed","paper":{"title":"Multi-Agent AI Control: Distributed Attacks Hamper Per-Instance Monitors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.MA"],"primary_cat":"cs.LG","authors_text":"Mary Phuong, Oliver Makins, Orazio Angelini, Zohreh Shams","submitted_at":"2026-07-08T13:03:25Z","abstract_excerpt":"AI control is a family of techniques to prevent an AI with malicious goals from subverting its operator's intent. AI Control usually studies a single agent in one trajectory, but real deployments run many agents over shared infrastructure, and the most severe risks (model-weight exfiltration, training-run poisoning) plausibly need several agents acting in concert. We initiate the empirical study of multi-agent AI control, formalising distributed attacks in which several agents jointly aim for a malicious goal. We develop FakeLab: a synthetic AI-lab codebase (9 services, 86 benign tasks, 4 atta"},"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":"2607.07368","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-08T13:03:25Z","cross_cats_sorted":["cs.AI","cs.MA"],"title_canon_sha256":"ccd60c1dcea8457b2032487706f797102e3414149e5250cda0341cd3dde37639","abstract_canon_sha256":"e5ad1a06b7f2e519bf2969bbaac9754d4c69bcbdd358311de5d0c6fd0bd1e3a6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-09T01:20:22.801526Z","signature_b64":"089XuPmh7djJJ7nx6x/5nilfnAspmbjJqzhfKz9Uz1Co6YNcmqI5SKDmCGaUojUem+3PG3jPfLCLiKG0PJy8BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0f6caeb2a2a18cd5e467a4cbe87e21a595bdd798b87fc43750dfed097ac01b2b","last_reissued_at":"2026-07-09T01:20:22.801134Z","signature_status":"signed_v1","first_computed_at":"2026-07-09T01:20:22.801134Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Multi-Agent AI Control: Distributed Attacks Hamper Per-Instance Monitors","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.MA"],"primary_cat":"cs.LG","authors_text":"Mary Phuong, Oliver Makins, Orazio Angelini, Zohreh Shams","submitted_at":"2026-07-08T13:03:25Z","abstract_excerpt":"AI control is a family of techniques to prevent an AI with malicious goals from subverting its operator's intent. AI Control usually studies a single agent in one trajectory, but real deployments run many agents over shared infrastructure, and the most severe risks (model-weight exfiltration, training-run poisoning) plausibly need several agents acting in concert. We initiate the empirical study of multi-agent AI control, formalising distributed attacks in which several agents jointly aim for a malicious goal. We develop FakeLab: a synthetic AI-lab codebase (9 services, 86 benign tasks, 4 atta"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.07368","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/2607.07368/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":"2607.07368","created_at":"2026-07-09T01:20:22.801186+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.07368v1","created_at":"2026-07-09T01:20:22.801186+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.07368","created_at":"2026-07-09T01:20:22.801186+00:00"},{"alias_kind":"pith_short_12","alias_value":"B5WK5MVCUGGN","created_at":"2026-07-09T01:20:22.801186+00:00"},{"alias_kind":"pith_short_16","alias_value":"B5WK5MVCUGGNLZDH","created_at":"2026-07-09T01:20:22.801186+00:00"},{"alias_kind":"pith_short_8","alias_value":"B5WK5MVC","created_at":"2026-07-09T01:20:22.801186+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/B5WK5MVCUGGNLZDHUTF6Q7RBUW","json":"https://pith.science/pith/B5WK5MVCUGGNLZDHUTF6Q7RBUW.json","graph_json":"https://pith.science/api/pith-number/B5WK5MVCUGGNLZDHUTF6Q7RBUW/graph.json","events_json":"https://pith.science/api/pith-number/B5WK5MVCUGGNLZDHUTF6Q7RBUW/events.json","paper":"https://pith.science/paper/B5WK5MVC"},"agent_actions":{"view_html":"https://pith.science/pith/B5WK5MVCUGGNLZDHUTF6Q7RBUW","download_json":"https://pith.science/pith/B5WK5MVCUGGNLZDHUTF6Q7RBUW.json","view_paper":"https://pith.science/paper/B5WK5MVC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.07368&json=true","fetch_graph":"https://pith.science/api/pith-number/B5WK5MVCUGGNLZDHUTF6Q7RBUW/graph.json","fetch_events":"https://pith.science/api/pith-number/B5WK5MVCUGGNLZDHUTF6Q7RBUW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/B5WK5MVCUGGNLZDHUTF6Q7RBUW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/B5WK5MVCUGGNLZDHUTF6Q7RBUW/action/storage_attestation","attest_author":"https://pith.science/pith/B5WK5MVCUGGNLZDHUTF6Q7RBUW/action/author_attestation","sign_citation":"https://pith.science/pith/B5WK5MVCUGGNLZDHUTF6Q7RBUW/action/citation_signature","submit_replication":"https://pith.science/pith/B5WK5MVCUGGNLZDHUTF6Q7RBUW/action/replication_record"}},"created_at":"2026-07-09T01:20:22.801186+00:00","updated_at":"2026-07-09T01:20:22.801186+00:00"}