{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.21147", "kind": "paper", "url": "https://arxiv.org/abs/2606.21147"}], "authors": ["Jiaxi Yang", "Chaewan Chun", "Jason Lucas", "Yuchen Yang", "Dongwon Lee"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.21147", "source_url": "https://arxiv.org/abs/2606.21147"}, "categories": ["General AI", "Speech & Audio", "cs.SD"], "description": {"en": "AOR-Bench is a benchmark of 3,000 pseudo-harmful audio samples across six categories, designed to evaluate over-refusal in large audio language models. It assesses whether models incorrectly reject benign audio queries."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.21147", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.21147", "title": "AOR-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.21147", "modality": null, "name": "AOR-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_aor-bench_86ee63c7", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "fe68c3461214133238bce2d73d1f0b445ba69dcdb819cfca05c2da60973148c2", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.21147"}, "publisher": null, "released": "2026-06-19", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.21147", "source_url": "https://arxiv.org/abs/2606.21147"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-21147", "source": "claire_radar", "source_benchmark_id": "2606.21147", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Speech & Audio", "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/2606.21147", "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": "a5a17e203a80a3d4a5cef9a08d3c84445573f948d1506a6a3925bb338ca90aff", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "AOR-Bench is a benchmark of 3,000 pseudo-harmful audio samples across six categories, designed to evaluate over-refusal in large audio language models. It assesses whether models incorrectly reject benign audio queries.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "To study this problem, we introduce \\textbf{AOR-Bench} (\\textbf{A}udio \\textbf{O}ver-\\textbf{R}efusal \\textbf{Bench}mark), the first benchmark for over-refusal specifically designed for LALMs."}, "familyId": "bmf_b04fbb60d247", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_aor-bench_86ee63c7", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.21147", "paper": "https://arxiv.org/abs/2606.21147", "pdf": "https://arxiv.org/pdf/2606.21147", "project": null, "report": "https://arxiv.org/abs/2606.21147"}, "metrics": [], "motivation": "Large Audio Language Models (LALMs) have demonstrated strong performance across a wide range of audio tasks.", "name": "AOR-Bench", "oneLine": "AOR-Bench is a benchmark of 3,000 pseudo-harmful audio samples across six categories, designed to evaluate over-refusal in large audio language models. It assesses whether models incorrectly reject benign audio queries.", "paperTitle": "AOR-Bench: Do Large Audio Language Models Over-Refuse Pseudo-Harmful Queries?", "primaryDomain": "General AI", "publication": {"evidence": "To appear in EMNLP 2026", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.21147", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "EMNLP 2026", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "AOR-Bench Team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://arxiv.org/abs/2606.21147"}], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-19", "source": {"authors": ["Jiaxi Yang", "Chaewan Chun", "Jason Lucas", "Yuchen Yang", "Dongwon Lee"], "categories": ["cs.SD", "cs.AI"], "id": "2606.21147", "title": "AOR-Bench: Do Large Audio Language Models Over-Refuse Pseudo-Harmful Queries?", "type": "arxiv", "url": "https://arxiv.org/abs/2606.21147"}, "sourceUpdatedAt": "2026-06-19T06:37:32Z", "topics": ["cs.SD"], "venueAttempts": [{"decisionRaw": "To appear in EMNLP 2026", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "To appear in EMNLP 2026", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.21147"}], "reviewStatus": "accepted", "venueName": "EMNLP 2026"}], "whyItMatters": "Over-refusal in audio models is understudied. AOR-Bench provides a standardized set to measure this behavior, helping align safety mechanisms with real-world audio context."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
