{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:OGBTINMGWGXAIV67SZUJQVALR5","short_pith_number":"pith:OGBTINMG","schema_version":"1.0","canonical_sha256":"7183343586b1ae0457df966898540b8f78e0cd0cb032fecf6111b5d38542fcbe","source":{"kind":"arxiv","id":"2404.13161","version":1},"attestation_state":"computed","paper":{"title":"CyberSecEval 2: A Wide-Ranging Cybersecurity Evaluation Suite for Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Cornelius Aschermann, Cyrus Nikolaidis, Daniel Song, David Molnar, Dhaval Kapil, Faizan Ahmad, Joshua Saxe, Manish Bhatt, Sahana Chennabasappa, Shengye Wan, Spencer Whitman, Yaohui Chen, Yue Li","submitted_at":"2024-04-19T20:11:12Z","abstract_excerpt":"Large language models (LLMs) introduce new security risks, but there are few comprehensive evaluation suites to measure and reduce these risks. We present BenchmarkName, a novel benchmark to quantify LLM security risks and capabilities. We introduce two new areas for testing: prompt injection and code interpreter abuse. We evaluated multiple state-of-the-art (SOTA) LLMs, including GPT-4, Mistral, Meta Llama 3 70B-Instruct, and Code Llama. Our results show that conditioning away risk of attack remains an unsolved problem; for example, all tested models showed between 26% and 41% successful prom"},"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":"2404.13161","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CR","submitted_at":"2024-04-19T20:11:12Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"da5682b79c6e4a5430e7a3cfbe661b6629bdddfcbe1ae8637b6f162d2ff1d54e","abstract_canon_sha256":"f60d3294ba59e41a5fc5c02d6fbc6cca8e8c533bfc81adad8f8596e628d73b95"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:10:28.295192Z","signature_b64":"VmbDRQ/rPosWG3SBxcIKlBmdGASxxdTlzvT5gFefNjeLGWcWs0BIS3t1YBebtJbzw0xgzV6Duj54dGRPT/opBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7183343586b1ae0457df966898540b8f78e0cd0cb032fecf6111b5d38542fcbe","last_reissued_at":"2026-07-05T08:10:28.294692Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:10:28.294692Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"CyberSecEval 2: A Wide-Ranging Cybersecurity Evaluation Suite for Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CR","authors_text":"Cornelius Aschermann, Cyrus Nikolaidis, Daniel Song, David Molnar, Dhaval Kapil, Faizan Ahmad, Joshua Saxe, Manish Bhatt, Sahana Chennabasappa, Shengye Wan, Spencer Whitman, Yaohui Chen, Yue Li","submitted_at":"2024-04-19T20:11:12Z","abstract_excerpt":"Large language models (LLMs) introduce new security risks, but there are few comprehensive evaluation suites to measure and reduce these risks. We present BenchmarkName, a novel benchmark to quantify LLM security risks and capabilities. We introduce two new areas for testing: prompt injection and code interpreter abuse. We evaluated multiple state-of-the-art (SOTA) LLMs, including GPT-4, Mistral, Meta Llama 3 70B-Instruct, and Code Llama. Our results show that conditioning away risk of attack remains an unsolved problem; for example, all tested models showed between 26% and 41% successful prom"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.13161","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/2404.13161/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":"2404.13161","created_at":"2026-07-05T08:10:28.294751+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.13161v1","created_at":"2026-07-05T08:10:28.294751+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.13161","created_at":"2026-07-05T08:10:28.294751+00:00"},{"alias_kind":"pith_short_12","alias_value":"OGBTINMGWGXA","created_at":"2026-07-05T08:10:28.294751+00:00"},{"alias_kind":"pith_short_16","alias_value":"OGBTINMGWGXAIV67","created_at":"2026-07-05T08:10:28.294751+00:00"},{"alias_kind":"pith_short_8","alias_value":"OGBTINMG","created_at":"2026-07-05T08:10:28.294751+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":22,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.05842","citing_title":"Beyond Refusal: A Same-Lineage Study of Aligned and Abliterated LLMs for Vulnerability Analysis","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2606.21389","citing_title":"From Production SIEM to Reusable Cybersecurity Artifacts","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23243","citing_title":"Are Frontier LLMs Ready for Cybersecurity? Evidence for Vertical Foundation Models from Dual-Mode Vulnerability Benchmarks","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.29981","citing_title":"Hephaestus: Toward a Cybersecurity AI Scientist","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22643","citing_title":"Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23243","citing_title":"Are Frontier LLMs Ready for Cybersecurity? Evidence for Vertical Foundation Models from Dual-Mode Vulnerability Benchmarks","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22643","citing_title":"Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety","ref_index":7,"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":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.19722","citing_title":"Measuring Safety Alignment Effects in Autonomous Security Agents","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17416","citing_title":"Benchmarking Mythos-Linked Bug Rediscovery","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16829","citing_title":"Constrained Code Generation with Discrete Diffusion","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2509.06921","citing_title":"Neuro-Symbolic AI for Cybersecurity: State of the Art, Challenges, and Opportunities","ref_index":177,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03081","citing_title":"Supply-Chain Poisoning Attacks Against LLM Coding Agent Skill Ecosystems","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11360","citing_title":"Options, Not Clicks: Lattice Refinement for Consent-Driven MCP Authorization","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09504","citing_title":"AI Security Policy Should Assess Systems, Not Only Models","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10597","citing_title":"CrackMeBench: Binary Reverse Engineering for Agents","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08316","citing_title":"AI-Driven Security Alert Screening and Alert Fatigue Mitigation in Security Operations Centers: A Comprehensive Survey","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05501","citing_title":"SOCpilot: Verifying Policy Compliance for LLM-Assisted Incident Response","ref_index":34,"is_internal_anchor":false},{"citing_arxiv_id":"2605.00081","citing_title":"Alignment Contracts for Agentic Security Systems","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18718","citing_title":"Towards Optimal Agentic Architectures for Offensive Security Tasks","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07830","citing_title":"CyBiasBench: Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios","ref_index":2,"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":48,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5","json":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5.json","graph_json":"https://pith.science/api/pith-number/OGBTINMGWGXAIV67SZUJQVALR5/graph.json","events_json":"https://pith.science/api/pith-number/OGBTINMGWGXAIV67SZUJQVALR5/events.json","paper":"https://pith.science/paper/OGBTINMG"},"agent_actions":{"view_html":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5","download_json":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5.json","view_paper":"https://pith.science/paper/OGBTINMG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.13161&json=true","fetch_graph":"https://pith.science/api/pith-number/OGBTINMGWGXAIV67SZUJQVALR5/graph.json","fetch_events":"https://pith.science/api/pith-number/OGBTINMGWGXAIV67SZUJQVALR5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5/action/storage_attestation","attest_author":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5/action/author_attestation","sign_citation":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5/action/citation_signature","submit_replication":"https://pith.science/pith/OGBTINMGWGXAIV67SZUJQVALR5/action/replication_record"}},"created_at":"2026-07-05T08:10:28.294751+00:00","updated_at":"2026-07-05T08:10:28.294751+00:00"}