{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:646EFKCIIWOCX6PTUIL5OGAEQR","short_pith_number":"pith:646EFKCI","schema_version":"1.0","canonical_sha256":"f73c42a848459c2bf9f3a217d71804847e3718fc288b383e30cb3c421104ddec","source":{"kind":"arxiv","id":"2508.19461","version":1},"attestation_state":"computed","paper":{"title":"Reliable Weak-to-Strong Monitoring of LLM Agents","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.AI","authors_text":"Ankit Aich, Chen Bo Calvin Zhang, Christina Q. Knight, Kevin Zhu, Neil Kale, Paula Rodriguez, Scale Red Team, Zifan Wang","submitted_at":"2025-08-26T22:29:31Z","abstract_excerpt":"We stress test monitoring systems for detecting covert misbehavior in autonomous LLM agents (e.g., secretly sharing private information). To this end, we systematize a monitor red teaming (MRT) workflow that incorporates: (1) varying levels of agent and monitor situational awareness; (2) distinct adversarial strategies to evade the monitor, such as prompt injection; and (3) two datasets and environments -- SHADE-Arena for tool-calling agents and our new CUA-SHADE-Arena, which extends TheAgentCompany, for computer-use agents. We run MRT on existing LLM monitor scaffoldings, which orchestrate LL"},"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":"2508.19461","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2025-08-26T22:29:31Z","cross_cats_sorted":["cs.CR","cs.LG"],"title_canon_sha256":"643729ae8ab8b5edce59204467db34360a37c62cd89453a04d397da02439851f","abstract_canon_sha256":"c9af5a2efc49d71baea690d8516f6114c9fbc36c710e0d12663cc2fe29623263"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:00:11.923254Z","signature_b64":"zm0GXbPaGGzrxpIsJiyCZMRNmwY8KcKsPCxwdGTI+0VxIxT7UpMYXxyOxwG69v3bmkdAf5r/9tgeYm/rAgPZAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f73c42a848459c2bf9f3a217d71804847e3718fc288b383e30cb3c421104ddec","last_reissued_at":"2026-07-05T12:00:11.922738Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:00:11.922738Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Reliable Weak-to-Strong Monitoring of LLM Agents","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CR","cs.LG"],"primary_cat":"cs.AI","authors_text":"Ankit Aich, Chen Bo Calvin Zhang, Christina Q. Knight, Kevin Zhu, Neil Kale, Paula Rodriguez, Scale Red Team, Zifan Wang","submitted_at":"2025-08-26T22:29:31Z","abstract_excerpt":"We stress test monitoring systems for detecting covert misbehavior in autonomous LLM agents (e.g., secretly sharing private information). To this end, we systematize a monitor red teaming (MRT) workflow that incorporates: (1) varying levels of agent and monitor situational awareness; (2) distinct adversarial strategies to evade the monitor, such as prompt injection; and (3) two datasets and environments -- SHADE-Arena for tool-calling agents and our new CUA-SHADE-Arena, which extends TheAgentCompany, for computer-use agents. We run MRT on existing LLM monitor scaffoldings, which orchestrate LL"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.19461","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/2508.19461/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":"2508.19461","created_at":"2026-07-05T12:00:11.922801+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.19461v1","created_at":"2026-07-05T12:00:11.922801+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.19461","created_at":"2026-07-05T12:00:11.922801+00:00"},{"alias_kind":"pith_short_12","alias_value":"646EFKCIIWOC","created_at":"2026-07-05T12:00:11.922801+00:00"},{"alias_kind":"pith_short_16","alias_value":"646EFKCIIWOCX6PT","created_at":"2026-07-05T12:00:11.922801+00:00"},{"alias_kind":"pith_short_8","alias_value":"646EFKCI","created_at":"2026-07-05T12:00:11.922801+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.05647","citing_title":"Coding with \"Enemy\": Can Human Developers Detect AI Agent Sabotage?","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15377","citing_title":"Ensemble Monitoring for AI Control: Diverse Signals Outweigh More Compute","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08715","citing_title":"AgentForesight: Online Auditing for Early Failure Prediction in Multi-Agent Systems","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12746","citing_title":"CoT-Guard: Small Models for Strong Monitoring","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08715","citing_title":"AgentForesight: Online Auditing for Early Failure Prediction in Multi-Agent Systems","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR","json":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR.json","graph_json":"https://pith.science/api/pith-number/646EFKCIIWOCX6PTUIL5OGAEQR/graph.json","events_json":"https://pith.science/api/pith-number/646EFKCIIWOCX6PTUIL5OGAEQR/events.json","paper":"https://pith.science/paper/646EFKCI"},"agent_actions":{"view_html":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR","download_json":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR.json","view_paper":"https://pith.science/paper/646EFKCI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.19461&json=true","fetch_graph":"https://pith.science/api/pith-number/646EFKCIIWOCX6PTUIL5OGAEQR/graph.json","fetch_events":"https://pith.science/api/pith-number/646EFKCIIWOCX6PTUIL5OGAEQR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR/action/storage_attestation","attest_author":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR/action/author_attestation","sign_citation":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR/action/citation_signature","submit_replication":"https://pith.science/pith/646EFKCIIWOCX6PTUIL5OGAEQR/action/replication_record"}},"created_at":"2026-07-05T12:00:11.922801+00:00","updated_at":"2026-07-05T12:00:11.922801+00:00"}