{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XM7PW25HTNXLYYX5VM6J3PGTKW","short_pith_number":"pith:XM7PW25H","schema_version":"1.0","canonical_sha256":"bb3efb6ba79b6ebc62fdab3c9dbcd3558528b990cf736c4c75c4e44b778a2554","source":{"kind":"arxiv","id":"2506.03053","version":2},"attestation_state":"computed","paper":{"title":"MAEBE: Multi-Agent Emergent Behavior Framework","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CY","cs.LG"],"primary_cat":"cs.MA","authors_text":"Martin Leitgab (Independent Researcher), Ram Potham (Independent Researcher), Sinem Erisken (Independent Researcher), Timothy Gothard (Independent Researcher)","submitted_at":"2025-06-03T16:33:47Z","abstract_excerpt":"Traditional AI safety evaluations on isolated LLMs are insufficient as multi-agent AI ensembles become prevalent, introducing novel emergent risks. This paper introduces the Multi-Agent Emergent Behavior Evaluation (MAEBE) framework to systematically assess such risks. Using MAEBE with the Greatest Good Benchmark (and a novel double-inversion question technique), we demonstrate that: (1) LLM moral preferences, particularly for Instrumental Harm, are surprisingly brittle and shift significantly with question framing, both in single agents and ensembles. (2) The moral reasoning of LLM ensembles "},"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":"2506.03053","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.MA","submitted_at":"2025-06-03T16:33:47Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CY","cs.LG"],"title_canon_sha256":"baeae7aee83866003aa246d176d34546fd97f4df57e82a5d6cf60897c6cef186","abstract_canon_sha256":"96782dc38b3cf9120ab90349b2340e58b835fd88eacb531c3ac73e4aaa6bd684"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:34:51.892731Z","signature_b64":"VmZs8UJEXznEu9V0CkpNGW8q68ifQ/BOzYLK7z/ECNKWPciYR44WwZV/XuNnXQvPeQ2CaMVecsKxdEHyr3/FDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bb3efb6ba79b6ebc62fdab3c9dbcd3558528b990cf736c4c75c4e44b778a2554","last_reissued_at":"2026-07-05T11:34:51.892205Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:34:51.892205Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MAEBE: Multi-Agent Emergent Behavior Framework","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CY","cs.LG"],"primary_cat":"cs.MA","authors_text":"Martin Leitgab (Independent Researcher), Ram Potham (Independent Researcher), Sinem Erisken (Independent Researcher), Timothy Gothard (Independent Researcher)","submitted_at":"2025-06-03T16:33:47Z","abstract_excerpt":"Traditional AI safety evaluations on isolated LLMs are insufficient as multi-agent AI ensembles become prevalent, introducing novel emergent risks. This paper introduces the Multi-Agent Emergent Behavior Evaluation (MAEBE) framework to systematically assess such risks. Using MAEBE with the Greatest Good Benchmark (and a novel double-inversion question technique), we demonstrate that: (1) LLM moral preferences, particularly for Instrumental Harm, are surprisingly brittle and shift significantly with question framing, both in single agents and ensembles. (2) The moral reasoning of LLM ensembles "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.03053","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/2506.03053/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":"2506.03053","created_at":"2026-07-05T11:34:51.892266+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.03053v2","created_at":"2026-07-05T11:34:51.892266+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.03053","created_at":"2026-07-05T11:34:51.892266+00:00"},{"alias_kind":"pith_short_12","alias_value":"XM7PW25HTNXL","created_at":"2026-07-05T11:34:51.892266+00:00"},{"alias_kind":"pith_short_16","alias_value":"XM7PW25HTNXLYYX5","created_at":"2026-07-05T11:34:51.892266+00:00"},{"alias_kind":"pith_short_8","alias_value":"XM7PW25H","created_at":"2026-07-05T11:34:51.892266+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28235","citing_title":"Govern the Repository, Not the Agent: Measuring Ecosystem-Level Risk in AI-Native Software","ref_index":34,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW","json":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW.json","graph_json":"https://pith.science/api/pith-number/XM7PW25HTNXLYYX5VM6J3PGTKW/graph.json","events_json":"https://pith.science/api/pith-number/XM7PW25HTNXLYYX5VM6J3PGTKW/events.json","paper":"https://pith.science/paper/XM7PW25H"},"agent_actions":{"view_html":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW","download_json":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW.json","view_paper":"https://pith.science/paper/XM7PW25H","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.03053&json=true","fetch_graph":"https://pith.science/api/pith-number/XM7PW25HTNXLYYX5VM6J3PGTKW/graph.json","fetch_events":"https://pith.science/api/pith-number/XM7PW25HTNXLYYX5VM6J3PGTKW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW/action/storage_attestation","attest_author":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW/action/author_attestation","sign_citation":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW/action/citation_signature","submit_replication":"https://pith.science/pith/XM7PW25HTNXLYYX5VM6J3PGTKW/action/replication_record"}},"created_at":"2026-07-05T11:34:51.892266+00:00","updated_at":"2026-07-05T11:34:51.892266+00:00"}