{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:34SA2LA5ZRO6NVMZ4OYHTAFCDL","short_pith_number":"pith:34SA2LA5","schema_version":"1.0","canonical_sha256":"df240d2c1dcc5de6d599e3b07980a21af197b5d7e216443778fc25cabda099aa","source":{"kind":"arxiv","id":"2408.01605","version":2},"attestation_state":"computed","paper":{"title":"CYBERSECEVAL 3: Advancing the Evaluation of Cybersecurity Risks and Capabilities in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Cyrus Nikolaidis, Daniel Song, David Molnar, James Crnkovich, Jayson Grace, Joshua Saxe, Manish Bhatt, Sahana Chennabasappa, Shengye Wan, Spencer Whitman, Stephanie Ding, Vlad Ionescu, Yue Li","submitted_at":"2024-08-02T23:47:27Z","abstract_excerpt":"We are releasing a new suite of security benchmarks for LLMs, CYBERSECEVAL 3, to continue the conversation on empirically measuring LLM cybersecurity risks and capabilities. CYBERSECEVAL 3 assesses 8 different risks across two broad categories: risk to third parties, and risk to application developers and end users. Compared to previous work, we add new areas focused on offensive security capabilities: automated social engineering, scaling manual offensive cyber operations, and autonomous offensive cyber operations. In this paper we discuss applying these benchmarks to the Llama 3 models and a"},"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":"2408.01605","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2024-08-02T23:47:27Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"ce7f6708417735867eeb794e409f43942e53de4b1ffae67f569a76bd15a34198","abstract_canon_sha256":"4997b1efa3e3bc3cd79ff62df672fceab818c32a1c3c7b84c1e82f58ea17fc2a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:04:11.457136Z","signature_b64":"Lw1a/YkOJUaFvmmcwENzAerVgxUSXJcUCiF0DsEDIW/YJIPEwdnBq7v9N/S53zEyOByHRkHlE3sJe68y0g16Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"df240d2c1dcc5de6d599e3b07980a21af197b5d7e216443778fc25cabda099aa","last_reissued_at":"2026-07-05T09:04:11.456610Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:04:11.456610Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CYBERSECEVAL 3: Advancing the Evaluation of Cybersecurity Risks and Capabilities in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Cyrus Nikolaidis, Daniel Song, David Molnar, James Crnkovich, Jayson Grace, Joshua Saxe, Manish Bhatt, Sahana Chennabasappa, Shengye Wan, Spencer Whitman, Stephanie Ding, Vlad Ionescu, Yue Li","submitted_at":"2024-08-02T23:47:27Z","abstract_excerpt":"We are releasing a new suite of security benchmarks for LLMs, CYBERSECEVAL 3, to continue the conversation on empirically measuring LLM cybersecurity risks and capabilities. CYBERSECEVAL 3 assesses 8 different risks across two broad categories: risk to third parties, and risk to application developers and end users. Compared to previous work, we add new areas focused on offensive security capabilities: automated social engineering, scaling manual offensive cyber operations, and autonomous offensive cyber operations. In this paper we discuss applying these benchmarks to the Llama 3 models and a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.01605","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/2408.01605/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":"2408.01605","created_at":"2026-07-05T09:04:11.456669+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.01605v2","created_at":"2026-07-05T09:04:11.456669+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.01605","created_at":"2026-07-05T09:04:11.456669+00:00"},{"alias_kind":"pith_short_12","alias_value":"34SA2LA5ZRO6","created_at":"2026-07-05T09:04:11.456669+00:00"},{"alias_kind":"pith_short_16","alias_value":"34SA2LA5ZRO6NVMZ","created_at":"2026-07-05T09:04:11.456669+00:00"},{"alias_kind":"pith_short_8","alias_value":"34SA2LA5","created_at":"2026-07-05T09:04:11.456669+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2607.07109","citing_title":"Certifying Ghosts: How Cybersecurity AI Agents Break the EU Cyber Resilience Act","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2607.05842","citing_title":"Beyond Refusal: A Same-Lineage Study of Aligned and Abliterated LLMs for Vulnerability Analysis","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2606.26377","citing_title":"Verifying Intent and Harm: A Unified Defense Against LLM-Generated Threats","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03811","citing_title":"AI Agents Enable Adaptive Computer Worms","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29981","citing_title":"Hephaestus: Toward a Cybersecurity AI Scientist","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20351","citing_title":"Refusal Evaluation in Coding LLMs and Code Agents: A Systematic Review of Thirteen Malicious-Code Prompt Corpora (2023-2025)","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19722","citing_title":"Measuring Safety Alignment Effects in Autonomous Security Agents","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2602.05523","citing_title":"Capture the Flags: Family-Based Evaluation of Agentic LLMs via Semantics-Preserving Transformations","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02574","citing_title":"Understanding the Effects of Safety Unalignment on Large Language Models","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08382","citing_title":"SecureForge: Finding and Preventing Vulnerabilities in LLM-Generated Code via Prompt Optimization","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24966","citing_title":"Risk Reporting for Developers' Internal AI Model Use","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00081","citing_title":"Alignment Contracts for Agentic Security Systems","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05872","citing_title":"Swiss-Bench 003: Evaluating LLM Reliability and Adversarial Security for Swiss Regulatory Contexts","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20389","citing_title":"CyberCertBench: Evaluating LLMs in Cybersecurity Certification Knowledge","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03179","citing_title":"A Validated Prompt Bank for Malicious Code Generation: Separating Executable Weapons from Security Knowledge in 1,554 Consensus-Labeled Prompts","ref_index":49,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL","json":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL.json","graph_json":"https://pith.science/api/pith-number/34SA2LA5ZRO6NVMZ4OYHTAFCDL/graph.json","events_json":"https://pith.science/api/pith-number/34SA2LA5ZRO6NVMZ4OYHTAFCDL/events.json","paper":"https://pith.science/paper/34SA2LA5"},"agent_actions":{"view_html":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL","download_json":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL.json","view_paper":"https://pith.science/paper/34SA2LA5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.01605&json=true","fetch_graph":"https://pith.science/api/pith-number/34SA2LA5ZRO6NVMZ4OYHTAFCDL/graph.json","fetch_events":"https://pith.science/api/pith-number/34SA2LA5ZRO6NVMZ4OYHTAFCDL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL/action/storage_attestation","attest_author":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL/action/author_attestation","sign_citation":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL/action/citation_signature","submit_replication":"https://pith.science/pith/34SA2LA5ZRO6NVMZ4OYHTAFCDL/action/replication_record"}},"created_at":"2026-07-05T09:04:11.456669+00:00","updated_at":"2026-07-05T09:04:11.456669+00:00"}