{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.20663", "kind": "paper", "url": "https://arxiv.org/abs/2606.20663"}], "authors": ["Guido Freire", "Agustín Martínez-Suñé", "Viviana Cotik"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.20663", "source_url": "https://arxiv.org/abs/2606.20663"}, "categories": ["Biology & Drug Discovery", "General AI", "Language & Knowledge", "cs.AI"], "description": {"en": "DrugBench evaluates AI control protocols for mitigating medication-related harm using 3,671 medical conversations and FDA drug labels, covering drug interactions, contraindications, dosing constraints, and patient action restrictions. Introduces severity-based monitoring."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.20663", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.20663", "title": "DrugBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Universidad de Buenos Aires"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.20663", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.20663", "source_url": "https://openalex.org/W7165635366"}, "key": "claire-radar:2606.20663", "modality": null, "name": "DrugBench", "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_drugbench_dc75a792", "origin_source": "arxiv", "recognition_confidence": "0.85", "record_sha256": "47627696d37ddb4df28ab2e71d68ad4060365eb28a1190327b0398357c79f176", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.20663"}, "publisher": null, "released": "2026-06-10", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.20663", "source_url": "https://arxiv.org/abs/2606.20663"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-20663", "source": "claire_radar", "source_benchmark_id": "2606.20663", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI", "Biology & Drug Discovery"], "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/2606.20663", "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": "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": "175b15a623efd90db47501eae3eb494c23dd635f8d448b835c1d81e7434ba9e6", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "DrugBench evaluates AI control protocols for mitigating medication-related harm using 3,671 medical conversations and FDA drug labels, covering drug interactions, contraindications, dosing constraints, and patient action restrictions. Introduces severity-based monitoring.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence"], "snippet": "To this end, we introduce DrugBench, an AI control evaluation benchmark which combines 3,671 multi-turn medical conversations from HealthBench with drug information from official FDA labels, covering four categories of medication-related harm: drug interactions, contraindications, dosing constraints, and patient action restrictions."}, "familyId": "bmf_e6b83865905a", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_drugbench_dc75a792", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Pharma & Biotech"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.20663", "paper": "https://arxiv.org/abs/2606.20663", "pdf": "https://arxiv.org/pdf/2606.20663", "project": null, "report": "https://arxiv.org/abs/2606.20663"}, "metrics": [], "motivation": "Large Language Models have the potential to expand and improve the access to clinical information by enabling new ways of interacting with medical knowledge in natural language.", "name": "DrugBench", "oneLine": "DrugBench evaluates AI control protocols for mitigating medication-related harm using 3,671 medical conversations and FDA drug labels, covering drug interactions, contraindications, dosing constraints, and patient action restrictions. Introduces severity-based monitoring.", "paperTitle": "DrugBench: Evaluating AI Control Protocols for Medication Harm Mitigation", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.20663", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 0.85, "relation": "introduces", "releasedAt": "2026-06-10", "source": {"authors": ["Guido Freire", "Agust\\'in Mart\\'inez-Su\\~n\\'e and Viviana Cotik"], "categories": ["cs.AI", "cs.CL", "cs.LG"], "id": "2606.20663", "title": "DrugBench: Evaluating AI Control Protocols for Medication Harm Mitigation", "type": "arxiv", "url": "https://arxiv.org/abs/2606.20663"}, "sourceUpdatedAt": "2026-06-10T01:16:15Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "Addresses the safety-critical need to evaluate external safeguards for LLMs in medical QA, beyond simple accuracy. Useful for developers of safe medical AI systems and control protocols."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
