{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.00107", "kind": "paper", "url": "https://arxiv.org/abs/2607.00107"}], "authors": ["Saif Mahmud", "Fadul Sikder", "Yuede Ji", "Haotian Zhang", "Yu Lei"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.00107", "source_url": "https://arxiv.org/abs/2607.00107"}, "categories": ["Code generation", "Cybersecurity", "Safety", "Safety & Trustworthiness"], "description": {"en": "VULBENCH-CPP evaluates the safety of AI-generated C++ code using multi-tier verification including functional testing, static analysis, dynamic analysis, and bounded model checking. It includes 8,918 programs from three LLMs and human authors."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.00107", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.00107", "title": "VULBENCH-CPP"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["The University of Texas at Arlington"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2607.00107", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2607.00107", "source_url": "https://openalex.org/W7167053570"}, "key": "claire-radar:2607.00107", "modality": null, "name": "VULBENCH-CPP", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_vulbench-cpp_84620b34", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "7a232e0f4ca36bca8bd8251f6b304be5d643ff2f83d8670a9b1cd919726a6d69", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2607.00107"}, "publisher": null, "released": "2026-06-30", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.00107", "source_url": "https://arxiv.org/abs/2607.00107"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-00107", "source": "claire_radar", "source_benchmark_id": "2607.00107", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Cybersecurity"], "area": "Safety & Trustworthiness", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2607.00107", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "state": "unavailable"}}, "status": "unavailable"}, "capabilities": ["Code generation"], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "261920b3c0aa5b0e431cd835e8c27561c77eadfbb2f834c9e0caaa2a72e39042", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "VULBENCH-CPP evaluates the safety of AI-generated C++ code using multi-tier verification including functional testing, static analysis, dynamic analysis, and bounded model checking. It includes 8,918 programs from three LLMs and human authors.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce VULBENCH-CPP, a benchmark of 8,918 C++ programs from three open-weight LLMs (Gemma 3 27B IT, LLaMA 3.3 70B Instruct, Qwen 2.5 Coder 32B Instruct) and human authors across 851 competitive-programming tasks."}, "familyId": "bmf_12ecf4f9c98a", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_vulbench-cpp_84620b34", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Cybersecurity"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2607.00107", "paper": "https://arxiv.org/abs/2607.00107", "pdf": "https://arxiv.org/pdf/2607.00107", "project": null, "report": "https://arxiv.org/abs/2607.00107"}, "metrics": [], "motivation": "As large language models (LLMs) are increasingly deployed for systems programming, their ability to generate secure C++ code, where a single memory-safety failure creates an exploitable vulnerability, remains a critical concern.", "name": "VULBENCH-CPP", "oneLine": "VULBENCH-CPP evaluates the safety of AI-generated C++ code using multi-tier verification including functional testing, static analysis, dynamic analysis, and bounded model checking. It includes 8,918 programs from three LLMs and human authors.", "paperTitle": "The Illusion of Safety: Multi-Tier Verification of AI vs. Human C++ Code", "primaryDomain": "Cybersecurity", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.00107", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-30", "source": {"authors": ["Saif Mahmud", "Fadul Sikder", "Yuede Ji", "Haotian Zhang", "and Yu Lei"], "categories": ["cs.SE"], "id": "2607.00107", "title": "The Illusion of Safety: Multi-Tier Verification of AI vs. Human C++ Code", "type": "arxiv", "url": "https://arxiv.org/abs/2607.00107"}, "sourceUpdatedAt": "2026-08-02T05:30:57Z", "topics": ["Safety"], "venueAttempts": [], "whyItMatters": "Security of AI-generated code is critical, but evaluations often rely on a single method. VULBENCH-CPP provides a comprehensive multi-tier benchmark that reveals AI code is more prone to runtime violations and that no single verification tier is sufficient."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
