{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:TU5VGOVTAFKYPSGAETRU5ZAVU7","short_pith_number":"pith:TU5VGOVT","schema_version":"1.0","canonical_sha256":"9d3b533ab3015587c8c024e34ee415a7c1cb9fb8502563e5cb778e738d5eab63","source":{"kind":"arxiv","id":"2304.03279","version":4},"attestation_state":"computed","paper":{"title":"Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmark","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CY"],"primary_cat":"cs.LG","authors_text":"Alexander Pan, Andy Zou, Dan Hendrycks, Hanlin Zhang, Jonathan Ng, Jun Shern Chan, Nathaniel Li, Scott Emmons, Steven Basart, Thomas Woodside","submitted_at":"2023-04-06T17:59:03Z","abstract_excerpt":"Artificial agents have traditionally been trained to maximize reward, which may incentivize power-seeking and deception, analogous to how next-token prediction in language models (LMs) may incentivize toxicity. So do agents naturally learn to be Machiavellian? And how do we measure these behaviors in general-purpose models such as GPT-4? Towards answering these questions, we introduce MACHIAVELLI, a benchmark of 134 Choose-Your-Own-Adventure games containing over half a million rich, diverse scenarios that center on social decision-making. Scenario labeling is automated with LMs, which are mor"},"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":"2304.03279","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-04-06T17:59:03Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CY"],"title_canon_sha256":"b42b0c60ea6bea2f6df4781087a313e03304e6aeef6a7b4c712deb0133528885","abstract_canon_sha256":"ac40fd8e0e8f1a7bed2123f13cb94d3660ee962254061c2aa5e793d1ee9392d3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:19:57.579574Z","signature_b64":"4q6PxD9Si2FN4nVtVLZSOvvev5Ay3N6XyiLd0vz2BqFkc6tqm2EL69TP4s8bjKK9qXLvFRUpF5MpGkOwrWudCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9d3b533ab3015587c8c024e34ee415a7c1cb9fb8502563e5cb778e738d5eab63","last_reissued_at":"2026-07-05T06:19:57.579047Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:19:57.579047Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Do the Rewards Justify the Means? Measuring Trade-Offs Between Rewards and Ethical Behavior in the MACHIAVELLI Benchmark","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CY"],"primary_cat":"cs.LG","authors_text":"Alexander Pan, Andy Zou, Dan Hendrycks, Hanlin Zhang, Jonathan Ng, Jun Shern Chan, Nathaniel Li, Scott Emmons, Steven Basart, Thomas Woodside","submitted_at":"2023-04-06T17:59:03Z","abstract_excerpt":"Artificial agents have traditionally been trained to maximize reward, which may incentivize power-seeking and deception, analogous to how next-token prediction in language models (LMs) may incentivize toxicity. So do agents naturally learn to be Machiavellian? And how do we measure these behaviors in general-purpose models such as GPT-4? Towards answering these questions, we introduce MACHIAVELLI, a benchmark of 134 Choose-Your-Own-Adventure games containing over half a million rich, diverse scenarios that center on social decision-making. Scenario labeling is automated with LMs, which are mor"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2304.03279","kind":"arxiv","version":4},"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/2304.03279/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":"2304.03279","created_at":"2026-07-05T06:19:57.579115+00:00"},{"alias_kind":"arxiv_version","alias_value":"2304.03279v4","created_at":"2026-07-05T06:19:57.579115+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2304.03279","created_at":"2026-07-05T06:19:57.579115+00:00"},{"alias_kind":"pith_short_12","alias_value":"TU5VGOVTAFKY","created_at":"2026-07-05T06:19:57.579115+00:00"},{"alias_kind":"pith_short_16","alias_value":"TU5VGOVTAFKYPSGA","created_at":"2026-07-05T06:19:57.579115+00:00"},{"alias_kind":"pith_short_8","alias_value":"TU5VGOVT","created_at":"2026-07-05T06:19:57.579115+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":11,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17478","citing_title":"Decoding Hidden Deception in Reasoning LLMs: Activation Explainers for Deception Auditing","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.07805","citing_title":"Beyond Goodhart's Law: A Dynamic Benchmark for Evaluating Compliance in Multi-Agent Systems","ref_index":56,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22643","citing_title":"Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2603.14987","citing_title":"Beyond Benchmark Islands: Toward Representative Trustworthiness Evaluation for Agentic AI","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22643","citing_title":"Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16282","citing_title":"Taxonomy and Consistency Analysis of Safety Benchmarks for AI Agents","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2511.08565","citing_title":"Moral Susceptibility and Robustness under Persona Role-Play in Large Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2402.05070","citing_title":"A Roadmap to Pluralistic Alignment","ref_index":190,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15340","citing_title":"Restoration, Exploration and Transformation: How Youth Engage Character.AI Chatbots for Feels, Fun and Finding themselves","ref_index":64,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10310","citing_title":"Positive Alignment: Artificial Intelligence for Human Flourishing","ref_index":148,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":228,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7","json":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7.json","graph_json":"https://pith.science/api/pith-number/TU5VGOVTAFKYPSGAETRU5ZAVU7/graph.json","events_json":"https://pith.science/api/pith-number/TU5VGOVTAFKYPSGAETRU5ZAVU7/events.json","paper":"https://pith.science/paper/TU5VGOVT"},"agent_actions":{"view_html":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7","download_json":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7.json","view_paper":"https://pith.science/paper/TU5VGOVT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2304.03279&json=true","fetch_graph":"https://pith.science/api/pith-number/TU5VGOVTAFKYPSGAETRU5ZAVU7/graph.json","fetch_events":"https://pith.science/api/pith-number/TU5VGOVTAFKYPSGAETRU5ZAVU7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7/action/storage_attestation","attest_author":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7/action/author_attestation","sign_citation":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7/action/citation_signature","submit_replication":"https://pith.science/pith/TU5VGOVTAFKYPSGAETRU5ZAVU7/action/replication_record"}},"created_at":"2026-07-05T06:19:57.579115+00:00","updated_at":"2026-07-05T06:19:57.579115+00:00"}