{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.04460", "kind": "paper", "url": "https://arxiv.org/abs/2606.04460"}], "authors": ["Tianneng Shi", "Robin Rheem", "Dongwei Jiang", "Mona Wang", "Francisco De La Riega", "Zhun Wang", "Jingzhi Jiang", "Alexander Cheung", "Sean Tai", "Jonah Cha", "Jianhong Tu", "Gabriel Han", "Chenguang Wang", "Jingxuan He", "Wenbo Guo", "Dawn Song"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.04460", "source_url": "https://arxiv.org/abs/2606.04460"}, "categories": ["Cybersecurity", "Language & Knowledge", "cs.CR"], "description": {"en": "CyberGym-E2E evaluates AI agents on end-to-end cybersecurity tasks, covering vulnerability discovery, proof-of-concept generation, and patch generation, using 920 real-world vulnerabilities from 139 open-source projects."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.04460", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.04460", "title": "CyberGym-E2E"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of California, Berkeley", "Johns Hopkins University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.04460", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.04460", "source_url": "https://openalex.org/W7163559119"}, "key": "claire-radar:2606.04460", "modality": null, "name": "CyberGym-E2E", "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_cybergym-e2e_b518f5fa", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "3163d89d2e12f662e2d01fd3a56dd21765a93805a3ee58744c3d5a2cbfa1d413", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.04460"}, "publisher": null, "released": "2026-06-03", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.04460", "source_url": "https://arxiv.org/abs/2606.04460"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-04460", "source": "claire_radar", "source_benchmark_id": "2606.04460", "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": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2606.04460", "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": [], "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": "77ad3eba7b618827dc512573d320f98e49e16da524e7a34a750085d1d2c611c8", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "CyberGym-E2E evaluates AI agents on end-to-end cybersecurity tasks, covering vulnerability discovery, proof-of-concept generation, and patch generation, using 920 real-world vulnerabilities from 139 open-source projects.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence"], "snippet": "To address this gap, we propose CyberGym-E2E, a large-scale and realistic end-to-end cybersecurity benchmark that comprehensively evaluates AI agents' abilities across the full lifecycle of vulnerability discovery, PoC generation, and patch generation."}, "familyId": "bmf_20d23ee666cc", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_cybergym-e2e_b518f5fa", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Cybersecurity"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.04460", "paper": "https://arxiv.org/abs/2606.04460", "pdf": "https://arxiv.org/pdf/2606.04460", "project": null, "report": "https://arxiv.org/abs/2606.04460"}, "metrics": [], "motivation": "AI has the potential to transform cybersecurity by enabling systems that can autonomously detect, analyze, and remediate software vulnerabilities.", "name": "CyberGym-E2E", "oneLine": "CyberGym-E2E evaluates AI agents on end-to-end cybersecurity tasks, covering vulnerability discovery, proof-of-concept generation, and patch generation, using 920 real-world vulnerabilities from 139 open-source projects.", "paperTitle": "CyberGym-E2E: Scalable Real-World Benchmark for AI Agents' End-to-End Cybersecurity Capabilities", "primaryDomain": "Cybersecurity", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.04460", "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": "2026-06-03", "source": {"authors": ["Tianneng Shi", "Robin Rheem", "Dongwei Jiang", "Mona Wang", "Francisco De La Riega", "Zhun Wang", "Jingzhi Jiang", "Alexander Cheung", "Sean Tai", "Jonah Cha", "Jianhong Tu", "Gabriel Han", "Chenguang Wang", "Jingxuan He", "Wenbo Guo", "Dawn Song"], "categories": ["cs.CR", "cs.AI", "cs.LG"], "id": "2606.04460", "title": "CyberGym-E2E: Scalable Real-World Benchmark for AI Agents' End-to-End Cybersecurity Capabilities", "type": "arxiv", "url": "https://arxiv.org/abs/2606.04460"}, "sourceUpdatedAt": "2026-07-17T23:45:25Z", "topics": ["cs.CR"], "venueAttempts": [], "whyItMatters": "Existing cybersecurity benchmarks often lack scale or end-to-end scope, limiting assessment of AI agents' practical utility in vulnerability remediation. CyberGym-E2E aims to address this by providing a large-scale, realistic environment for evaluating agents across the full vulnerability lifecycle."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
