{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:LF2DOOI5GWYHTRILVOYSNETQQN","short_pith_number":"pith:LF2DOOI5","schema_version":"1.0","canonical_sha256":"597437391d35b079c50babb1269270836931879273532ee628e43931d5bc4f52","source":{"kind":"arxiv","id":"2607.28685","version":1},"attestation_state":"computed","paper":{"title":"Safety, or Just Capability? A Validity Audit of Agent-Safety Benchmarks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.IR"],"primary_cat":"cs.AI","authors_text":"Bowen Liu, Dingyan Shang, Xiao Han, Youting Wang, Yuan Tang","submitted_at":"2026-07-30T03:45:10Z","abstract_excerpt":"Agent-safety benchmarks measure different behaviors, and their scores get quoted interchangeably as an agent's safety. We treat four of them (R-Judge, InjecAgent, AgentHarm, AgentDojo) as measurements to be validated, running each under its official implementation and author-provided scorer on up to 22 models, with MMLU and GPQA measured by us under one protocol as a capability composite. The metric is the first problem. On any binary trace-judgment benchmark scored by $F_1$, an ``always positive'' policy attains $F_1 = 2\\pi/(1+\\pi)$; on R-Judge that is $0.690$, above five of the 21 models tha"},"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":"2607.28685","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-30T03:45:10Z","cross_cats_sorted":["cs.IR"],"title_canon_sha256":"7d35204fc934013277a9c3601a371dd3def3b38301507a7fbc850b487ceccad0","abstract_canon_sha256":"09fa9955fcfe95fd1390941d4d8e6f2d886eea82069d9bd633fdf3cb6529ca76"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-03T00:12:10.579711Z","signature_b64":"yaY1sG9ZCWU/IqadIchO4avkeVdeLYvlP6OOcLHZL91pk/n3sRqm5Q3Z7Jc69CyS2ChBhrqqDZUybl8A+ygACw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"597437391d35b079c50babb1269270836931879273532ee628e43931d5bc4f52","last_reissued_at":"2026-08-03T00:12:10.577518Z","signature_status":"signed_v1","first_computed_at":"2026-08-03T00:12:10.577518Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Safety, or Just Capability? A Validity Audit of Agent-Safety Benchmarks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.IR"],"primary_cat":"cs.AI","authors_text":"Bowen Liu, Dingyan Shang, Xiao Han, Youting Wang, Yuan Tang","submitted_at":"2026-07-30T03:45:10Z","abstract_excerpt":"Agent-safety benchmarks measure different behaviors, and their scores get quoted interchangeably as an agent's safety. We treat four of them (R-Judge, InjecAgent, AgentHarm, AgentDojo) as measurements to be validated, running each under its official implementation and author-provided scorer on up to 22 models, with MMLU and GPQA measured by us under one protocol as a capability composite. The metric is the first problem. On any binary trace-judgment benchmark scored by $F_1$, an ``always positive'' policy attains $F_1 = 2\\pi/(1+\\pi)$; on R-Judge that is $0.690$, above five of the 21 models tha"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.28685","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/2607.28685/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":"2607.28685","created_at":"2026-08-03T00:12:10.579034+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.28685v1","created_at":"2026-08-03T00:12:10.579034+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.28685","created_at":"2026-08-03T00:12:10.579034+00:00"},{"alias_kind":"pith_short_12","alias_value":"LF2DOOI5GWYH","created_at":"2026-08-03T00:12:10.579034+00:00"},{"alias_kind":"pith_short_16","alias_value":"LF2DOOI5GWYHTRIL","created_at":"2026-08-03T00:12:10.579034+00:00"},{"alias_kind":"pith_short_8","alias_value":"LF2DOOI5","created_at":"2026-08-03T00:12:10.579034+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/LF2DOOI5GWYHTRILVOYSNETQQN","json":"https://pith.science/pith/LF2DOOI5GWYHTRILVOYSNETQQN.json","graph_json":"https://pith.science/api/pith-number/LF2DOOI5GWYHTRILVOYSNETQQN/graph.json","events_json":"https://pith.science/api/pith-number/LF2DOOI5GWYHTRILVOYSNETQQN/events.json","paper":"https://pith.science/paper/LF2DOOI5"},"agent_actions":{"view_html":"https://pith.science/pith/LF2DOOI5GWYHTRILVOYSNETQQN","download_json":"https://pith.science/pith/LF2DOOI5GWYHTRILVOYSNETQQN.json","view_paper":"https://pith.science/paper/LF2DOOI5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.28685&json=true","fetch_graph":"https://pith.science/api/pith-number/LF2DOOI5GWYHTRILVOYSNETQQN/graph.json","fetch_events":"https://pith.science/api/pith-number/LF2DOOI5GWYHTRILVOYSNETQQN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LF2DOOI5GWYHTRILVOYSNETQQN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LF2DOOI5GWYHTRILVOYSNETQQN/action/storage_attestation","attest_author":"https://pith.science/pith/LF2DOOI5GWYHTRILVOYSNETQQN/action/author_attestation","sign_citation":"https://pith.science/pith/LF2DOOI5GWYHTRILVOYSNETQQN/action/citation_signature","submit_replication":"https://pith.science/pith/LF2DOOI5GWYHTRILVOYSNETQQN/action/replication_record"}},"created_at":"2026-08-03T00:12:10.579034+00:00","updated_at":"2026-08-03T00:12:10.579034+00:00"}