{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.26548", "kind": "paper", "url": "https://arxiv.org/abs/2605.26548"}, {"id": "gh:sec-bench/sec-bench-pro", "kind": "repo", "url": "https://github.com/SEC-bench/SEC-bench-Pro"}], "authors": ["Hwiwon Lee", "Jiawei Liu", "Dongjun Kim", "Wubing Xia", "Ziqi Zhang", "Chunqiu Steven Xia", "Lingming Zhang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.26548", "source_url": "https://arxiv.org/abs/2605.26548"}, "categories": ["Cybersecurity", "Language & Knowledge", "cs.CR"], "description": {"en": "SEC-bench Pro measures long-horizon vulnerability discovery in software security. It includes 344 verified vulnerabilities across V8, SpiderMonkey, and Linux kernel, each paired with instructions for reproducing a working proof-of-concept. Grading uses an LLM-based judge to classify generated PoCs against vulnerable, fixed, and latest images."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.26548", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.26548", "title": "SEC-bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of Illinois Urbana-Champaign"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2605.26548", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2605.26548", "source_url": "https://openalex.org/W7162507072"}, "key": "claire-radar:2605.26548", "modality": null, "name": "SEC-bench", "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_sec-bench_29fdf2d8", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "e176031c8b11cd9d9ff59259107a2486f940fb695ecdeb4ea4038bea9c361642", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.26548"}, "publisher": null, "released": "2026-05-26", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.26548", "source_url": "https://arxiv.org/abs/2605.26548"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-26548", "source": "claire_radar", "source_benchmark_id": "2605.26548", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Cybersecurity"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/SEC-bench/SEC-bench-Pro", "githubScope": "benchmark_repo", "githubStars": 56, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2605.26548", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "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": "4a67b69f1e510b57e99bbf96d2c727261215a3f297a7c61b91228934e693307a", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SEC-bench Pro measures long-horizon vulnerability discovery in software security. It includes 344 verified vulnerabilities across V8, SpiderMonkey, and Linux kernel, each paired with instructions for reproducing a working proof-of-concept. Grading uses an LLM-based judge to classify generated PoCs against vulnerable, fixed, and latest images.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We present SEC-bench Pro, a benchmark that measures how well frontier models hunt real vulnerabilities by reproducing working PoC inputs from disclosed reports, where each task pairs a concrete bug with the instructions for triggering it."}, "familyId": "bmf_830a56b48245", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/SEC-bench/SEC-bench-Pro", "url": "https://github.com/SEC-bench/SEC-bench-Pro"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/SEC-bench/SEC-bench-Pro", "url": "https://huggingface.co/papers/2605.26548"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/SEC-bench/SEC-bench-Pro.", "url": "https://arxiv.org/abs/2605.26548"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/SEC-bench/SEC-bench-Pro", "url": "https://arxiv.org/abs/2605.26548"}], "scope": "benchmark_repo", "url": "https://github.com/SEC-bench/SEC-bench-Pro", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_sec-bench_29fdf2d8", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Cybersecurity"], "links": {"code": "https://github.com/SEC-bench/SEC-bench-Pro", "data": null, "hfPaper": "https://huggingface.co/papers/2605.26548", "paper": "https://arxiv.org/abs/2605.26548", "pdf": "https://arxiv.org/pdf/2605.26548", "project": null, "report": "https://arxiv.org/abs/2605.26548"}, "metrics": [], "motivation": "Finding a real vulnerability in complicated systems is a challenging, long-horizon task that demands reasoning across an entire codebase to produce a working proof-of-concept (PoC).", "name": "SEC-bench", "oneLine": "SEC-bench Pro measures long-horizon vulnerability discovery in software security. It includes 344 verified vulnerabilities across V8, SpiderMonkey, and Linux kernel, each paired with instructions for reproducing a working proof-of-concept. Grading uses an LLM-based judge to classify generated PoCs against vulnerable, fixed, and latest images.", "paperTitle": "SEC-bench Pro: Can Language Models Solve Long-Horizon Software Security Tasks?", "primaryDomain": "Cybersecurity", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.26548", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-26", "source": {"authors": ["Hwiwon Lee", "Jiawei Liu", "Dongjun Kim", "Wubing Xia", "Ziqi Zhang", "Chunqiu Steven Xia", "Lingming Zhang"], "categories": ["cs.CR", "cs.LG"], "id": "2605.26548", "title": "SEC-bench Pro: Can Language Models Solve Long-Horizon Software Security Tasks?", "type": "arxiv", "url": "https://arxiv.org/abs/2605.26548"}, "sourceUpdatedAt": "2026-07-20T16:24:12Z", "topics": ["cs.CR"], "venueAttempts": [], "whyItMatters": "Finding real vulnerabilities requires reasoning across an entire codebase to produce a working PoC, a challenging task that is understudied. SEC-bench Pro provides a reproducible environment for evaluating long-horizon security capabilities, helping identify where models succeed and fail in realistic vulnerability hunting."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
