{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.20982", "kind": "paper", "url": "https://arxiv.org/abs/2607.20982"}], "authors": ["Vishal Ishwar Naik", "Chenyu Xu", "Donna Dong", "Hussein Hassan", "Abhishek Pradhan", "Ofer Mendelevitch", "Tallat Shafat", "Humayun Irshad"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.20982", "source_url": "https://arxiv.org/abs/2607.20982"}, "categories": ["General AI", "Language & Knowledge", "cs.AI"], "description": {"en": "GuardianAgentBench (GABench) is a benchmark of 580 agent scenarios across six domains, evaluated on three production frameworks (LangChain, LlamaIndex, Vectara), with multi-stage validation and five adversarial attack modes to assess tool-use correctness and safety."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.20982", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.20982", "title": "GuardianAgentBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.20982", "modality": null, "name": "GuardianAgentBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "Medium", "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_guardianagentbench_89ff40f2", "origin_source": "arxiv", "recognition_confidence": "0.85", "record_sha256": "9d38e9fc364145aefaca6af7fe479bc508a0bf907608b5c1aa807d51eae971d7", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2607.20982"}, "publisher": null, "released": "2026-07-23", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.20982", "source_url": "https://arxiv.org/abs/2607.20982"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-20982", "source": "claire_radar", "source_benchmark_id": "2607.20982", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "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": 1, "hfPaperUrl": "https://huggingface.co/papers/2607.20982", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "969e7de6bcde2872f638de42b7d9814e7c538c18998495eee238c33023901d95", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "GuardianAgentBench (GABench) is a benchmark of 580 agent scenarios across six domains, evaluated on three production frameworks (LangChain, LlamaIndex, Vectara), with multi-stage validation and five adversarial attack modes to assess tool-use correctness and safety.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence"], "snippet": "We present GuardianAgentBench (GABench), a benchmark of 580 scenarios across six domains evaluated on three production-ready frameworks: LangChain, LlamaIndex, and Vectara."}, "familyId": "bmf_6f6ba219145f", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_guardianagentbench_89ff40f2", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2607.20982", "paper": "https://arxiv.org/abs/2607.20982", "pdf": "https://arxiv.org/pdf/2607.20982", "project": null, "report": "https://arxiv.org/abs/2607.20982"}, "metrics": [], "motivation": "As large language model agents increasingly operate autonomously with access to tools and external environments, ensuring their safe and reliable behavior becomes critical.", "name": "GuardianAgentBench", "oneLine": "GuardianAgentBench (GABench) is a benchmark of 580 agent scenarios across six domains, evaluated on three production frameworks (LangChain, LlamaIndex, Vectara), with multi-stage validation and five adversarial attack modes to assess tool-use correctness and safety.", "paperTitle": "GuardianAgentBench: Where Agents Fail and How to Guard Them", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.20982", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "GuardianAgentBench Team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://arxiv.org/abs/2607.20982"}], "ranking": {"90d": {"confidence": "Low", "coverage": 0.15, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": null, "value": null}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.12247537038455211, "value": 1}}, "confidence": "Low", "coverage": 0.15, "rank": null, "score": 44}, "method": "attention-ranking-v13", "rank": null, "score": 44}}, "readiness": "Paper only", "recognitionConfidence": 0.85, "relation": "introduces", "releasedAt": "2026-07-23", "source": {"authors": ["Vishal Ishwar Naik", "Chenyu Xu", "Donna Dong", "Hussein Hassan", "Abhishek Pradhan", "Ofer Mendelevitch", "Tallat Shafat and Humayun Irshad"], "categories": ["cs.AI"], "id": "2607.20982", "title": "GuardianAgentBench: Where Agents Fail and How to Guard Them", "type": "arxiv", "url": "https://arxiv.org/abs/2607.20982"}, "sourceUpdatedAt": "2026-07-23T07:05:05Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "LLM agents need structured evaluation of tool selection and failure modes; GABench provides a reusable framework with multiple attack modes and guardrail assessment, offering a fine-grained view of where agents fail."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
