{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:GKNDCVW2Q3PE6YYZ6D33GTO336","short_pith_number":"pith:GKNDCVW2","schema_version":"1.0","canonical_sha256":"329a3156da86de4f6319f0f7b34ddbdf8d01f530bee73ddfbdf867987ff950cf","source":{"kind":"arxiv","id":"2602.12316","version":2},"attestation_state":"computed","paper":{"title":"GT-HarmBench: Benchmarking AI Safety Risks Through the Lens of Game Theory","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CY","cs.GT","cs.MA"],"primary_cat":"cs.AI","authors_text":"Isabel Dahlgren, Pepijn Cobben, Terry Jingchen Zhang, Thao Amelia Pham, Xuanqiang Angelo Huang, Zhijing Jin","submitted_at":"2026-02-12T17:29:52Z","abstract_excerpt":"Frontier AI systems are increasingly capable and deployed in high-stakes multi-agent environments. However, existing AI safety benchmarks largely evaluate single agents, leaving multi-agent risks such as coordination failure and conflict poorly understood. We introduce GT-HarmBench, a benchmark of 1,535 high-stakes scenarios spanning game-theoretic structures such as the Prisoner's Dilemma, Stag Hunt and Chicken. Scenarios are drawn from realistic AI risk contexts in the MIT AI Risk Repository. Across 15 frontier models, agents fail to choose socially beneficial actions in 38% of high-stakes c"},"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":"2602.12316","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-02-12T17:29:52Z","cross_cats_sorted":["cs.CL","cs.CY","cs.GT","cs.MA"],"title_canon_sha256":"c459d4926902ceb164be4c5e35d959fe20ba3c2cb3f15e7faf49c9b71a3588ad","abstract_canon_sha256":"90b8a5d0b3507e98c62ab603ff3de2fd88c16c2c96b12b2b6c41213c7466c293"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:15.060424Z","signature_b64":"vPLoyuuTlf7XQDMgnU0icXumfoN2wgUATHcXFCjddRFuUYBz3Z+mUliE6ag4GDE+z7OmQznccg97ErxKMf+xCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"329a3156da86de4f6319f0f7b34ddbdf8d01f530bee73ddfbdf867987ff950cf","last_reissued_at":"2026-05-25T02:01:15.059572Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:15.059572Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GT-HarmBench: Benchmarking AI Safety Risks Through the Lens of Game Theory","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CY","cs.GT","cs.MA"],"primary_cat":"cs.AI","authors_text":"Isabel Dahlgren, Pepijn Cobben, Terry Jingchen Zhang, Thao Amelia Pham, Xuanqiang Angelo Huang, Zhijing Jin","submitted_at":"2026-02-12T17:29:52Z","abstract_excerpt":"Frontier AI systems are increasingly capable and deployed in high-stakes multi-agent environments. However, existing AI safety benchmarks largely evaluate single agents, leaving multi-agent risks such as coordination failure and conflict poorly understood. We introduce GT-HarmBench, a benchmark of 1,535 high-stakes scenarios spanning game-theoretic structures such as the Prisoner's Dilemma, Stag Hunt and Chicken. Scenarios are drawn from realistic AI risk contexts in the MIT AI Risk Repository. Across 15 frontier models, agents fail to choose socially beneficial actions in 38% of high-stakes c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.12316","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/2602.12316/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":"2602.12316","created_at":"2026-05-25T02:01:15.059693+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.12316v2","created_at":"2026-05-25T02:01:15.059693+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.12316","created_at":"2026-05-25T02:01:15.059693+00:00"},{"alias_kind":"pith_short_12","alias_value":"GKNDCVW2Q3PE","created_at":"2026-05-25T02:01:15.059693+00:00"},{"alias_kind":"pith_short_16","alias_value":"GKNDCVW2Q3PE6YYZ","created_at":"2026-05-25T02:01:15.059693+00:00"},{"alias_kind":"pith_short_8","alias_value":"GKNDCVW2","created_at":"2026-05-25T02:01:15.059693+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.17510","citing_title":"Scale-Dependent Collective Adaptation in Self-Amending LLM Societies: A Cross-Family Study of Emergent Governance","ref_index":15,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08426","citing_title":"Mechanism Design Is Not Enough: Prosocial Agents for Cooperative AI","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21282","citing_title":"Strategic Heterogeneous Multi-Agent Architecture for Cost-Effective Code Vulnerability Detection","ref_index":1,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336","json":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336.json","graph_json":"https://pith.science/api/pith-number/GKNDCVW2Q3PE6YYZ6D33GTO336/graph.json","events_json":"https://pith.science/api/pith-number/GKNDCVW2Q3PE6YYZ6D33GTO336/events.json","paper":"https://pith.science/paper/GKNDCVW2"},"agent_actions":{"view_html":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336","download_json":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336.json","view_paper":"https://pith.science/paper/GKNDCVW2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.12316&json=true","fetch_graph":"https://pith.science/api/pith-number/GKNDCVW2Q3PE6YYZ6D33GTO336/graph.json","fetch_events":"https://pith.science/api/pith-number/GKNDCVW2Q3PE6YYZ6D33GTO336/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336/action/storage_attestation","attest_author":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336/action/author_attestation","sign_citation":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336/action/citation_signature","submit_replication":"https://pith.science/pith/GKNDCVW2Q3PE6YYZ6D33GTO336/action/replication_record"}},"created_at":"2026-05-25T02:01:15.059693+00:00","updated_at":"2026-05-25T02:01:15.059693+00:00"}