{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TAYLZTS2HDEA5WUPNPQMUGXW2G","short_pith_number":"pith:TAYLZTS2","schema_version":"1.0","canonical_sha256":"9830bcce5a38c80eda8f6be0ca1af6d18b73d60240e40eca263b78ba685be7c9","source":{"kind":"arxiv","id":"2508.02921","version":1},"attestation_state":"computed","paper":{"title":"PentestJudge: Judging Agent Behavior Against Operational Requirements","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.AI","authors_text":"Max Harley, Michael Kouremetis, Shane Caldwell, Vincent Abruzzo, Will Pearce","submitted_at":"2025-08-04T21:52:50Z","abstract_excerpt":"We introduce PentestJudge, a system for evaluating the operations of penetration testing agents. PentestJudge is a large language model (LLM)-as-judge with access to tools that allow it to consume arbitrary trajectories of agent states and tool call history to determine whether a security agent's actions meet certain operating criteria that would be impractical to evaluate programmatically. We develop rubrics that use a tree structure to hierarchically collapse the penetration testing task for a particular environment into smaller, simpler, and more manageable sub-tasks and criteria until each"},"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":"2508.02921","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.AI","submitted_at":"2025-08-04T21:52:50Z","cross_cats_sorted":["cs.CR"],"title_canon_sha256":"e778842235cd32302f9df1858363a8af171ea86105ef0e83b4c072d0c2b9a9b7","abstract_canon_sha256":"f4329a531136a528ef8765f52d5a770cd48e433dcd9c183eecffc9b9ce305c4e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:48:32.914714Z","signature_b64":"IYRZzWb5rdpvbdsGjcizt5Can/Vw8NHHhgS7PcmxjrZE/a7NLlaH+MJY9Lv1jOQndAuiR1hNlT6TqLfS7eI7DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9830bcce5a38c80eda8f6be0ca1af6d18b73d60240e40eca263b78ba685be7c9","last_reissued_at":"2026-07-05T11:48:32.914291Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:48:32.914291Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PentestJudge: Judging Agent Behavior Against Operational Requirements","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.CR"],"primary_cat":"cs.AI","authors_text":"Max Harley, Michael Kouremetis, Shane Caldwell, Vincent Abruzzo, Will Pearce","submitted_at":"2025-08-04T21:52:50Z","abstract_excerpt":"We introduce PentestJudge, a system for evaluating the operations of penetration testing agents. PentestJudge is a large language model (LLM)-as-judge with access to tools that allow it to consume arbitrary trajectories of agent states and tool call history to determine whether a security agent's actions meet certain operating criteria that would be impractical to evaluate programmatically. We develop rubrics that use a tree structure to hierarchically collapse the penetration testing task for a particular environment into smaller, simpler, and more manageable sub-tasks and criteria until each"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.02921","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/2508.02921/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":"2508.02921","created_at":"2026-07-05T11:48:32.914353+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.02921v1","created_at":"2026-07-05T11:48:32.914353+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.02921","created_at":"2026-07-05T11:48:32.914353+00:00"},{"alias_kind":"pith_short_12","alias_value":"TAYLZTS2HDEA","created_at":"2026-07-05T11:48:32.914353+00:00"},{"alias_kind":"pith_short_16","alias_value":"TAYLZTS2HDEA5WUP","created_at":"2026-07-05T11:48:32.914353+00:00"},{"alias_kind":"pith_short_8","alias_value":"TAYLZTS2","created_at":"2026-07-05T11:48:32.914353+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07774","citing_title":"ScopeJudge: Cost-Aware Pre-Execution Gating for Offensive Security Agents","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10834","citing_title":"From Controlled to the Wild: Evaluation of Pentesting Agents for the Real-World","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04019","citing_title":"Redefining AI Red Teaming in the Agentic Era: From Weeks to Hours","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G","json":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G.json","graph_json":"https://pith.science/api/pith-number/TAYLZTS2HDEA5WUPNPQMUGXW2G/graph.json","events_json":"https://pith.science/api/pith-number/TAYLZTS2HDEA5WUPNPQMUGXW2G/events.json","paper":"https://pith.science/paper/TAYLZTS2"},"agent_actions":{"view_html":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G","download_json":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G.json","view_paper":"https://pith.science/paper/TAYLZTS2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.02921&json=true","fetch_graph":"https://pith.science/api/pith-number/TAYLZTS2HDEA5WUPNPQMUGXW2G/graph.json","fetch_events":"https://pith.science/api/pith-number/TAYLZTS2HDEA5WUPNPQMUGXW2G/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G/action/storage_attestation","attest_author":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G/action/author_attestation","sign_citation":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G/action/citation_signature","submit_replication":"https://pith.science/pith/TAYLZTS2HDEA5WUPNPQMUGXW2G/action/replication_record"}},"created_at":"2026-07-05T11:48:32.914353+00:00","updated_at":"2026-07-05T11:48:32.914353+00:00"}