{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.01317", "kind": "paper", "url": "https://arxiv.org/abs/2606.01317"}, {"id": "gh:sssr-lab/saber", "kind": "repo", "url": "https://github.com/sssr-lab/saber"}], "authors": ["Qi Hu", "Yifeng Tang", "Qinghua Wang", "Lanyang Zhao", "Pengji Zhang", "Yuhao Qing", "Xin Yao", "Dong Huang", "Lin Zhang", "Zhuoran Ji"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.01317", "source_url": "https://arxiv.org/abs/2606.01317"}, "categories": ["Agents", "Agents & Tool Use", "General AI", "Safety"], "description": {"en": "SABER evaluates operational safety of LLM coding agents in stateful project workspaces. Agents perform realistic tasks, and safety is scored from the final environment state after action sequences. Violations are categorized by cause, enabling model-specific safety profile analysis."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.01317", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.01317", "title": "SABER"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.01317", "modality": null, "name": "SABER", "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_saber_ac2e2d88", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "1b748ab205c305042e52d07ca940206598c9a97766fcd2c1e4d0630ecd6cfdbb", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.01317"}, "publisher": null, "released": "2026-05-31", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.01317", "source_url": "https://arxiv.org/abs/2606.01317"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-01317", "source": "claire_radar", "source_benchmark_id": "2606.01317", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Agents & Tool Use", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/sssr-lab/saber", "githubScope": "benchmark_repo", "githubStars": 7, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 2, "hfPaperUrl": "https://huggingface.co/papers/2606.01317", "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": "1e421050f57d308d4abc8168731c709ad0782aec2be5df5be06bd3189bfd7a0e", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SABER evaluates operational safety of LLM coding agents in stateful project workspaces. Agents perform realistic tasks, and safety is scored from the final environment state after action sequences. Violations are categorized by cause, enabling model-specific safety profile analysis.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "public artifact URL"], "snippet": "We present SABER, a benchmark for environment-aware operational safety that places models in realistic agent-style projects and evaluates safety from the final environment state after a sequence of actions."}, "familyId": "bmf_e4c3c153699b", "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/sssr-lab/saber", "url": "https://github.com/sssr-lab/saber"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/sssr-lab/saber", "url": "https://huggingface.co/papers/2606.01317"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/sssr-lab/saber", "url": "https://arxiv.org/abs/2606.01317"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/sssr-lab/saber.", "url": "https://arxiv.org/abs/2606.01317"}], "scope": "benchmark_repo", "url": "https://github.com/sssr-lab/saber", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_saber_ac2e2d88", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/sssr-lab/saber", "data": null, "hfPaper": "https://huggingface.co/papers/2606.01317", "paper": "https://arxiv.org/abs/2606.01317", "pdf": "https://arxiv.org/pdf/2606.01317", "project": null, "report": "https://arxiv.org/abs/2606.01317"}, "metrics": [], "motivation": "Large language models are increasingly deployed as coding agents, shifting safety from individual responses to action sequences.", "name": "SABER", "oneLine": "SABER evaluates operational safety of LLM coding agents in stateful project workspaces. Agents perform realistic tasks, and safety is scored from the final environment state after action sequences. Violations are categorized by cause, enabling model-specific safety profile analysis.", "paperTitle": "SABER: Benchmarking Operational Safety of LLM Coding Agents in Stateful Project Workspaces", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.01317", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "SSSR Lab", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/sssr-lab/saber"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-31", "source": {"authors": ["Qi Hu", "Yifeng Tang", "Qinghua Wang", "Lanyang Zhao", "Pengji Zhang", "Yuhao Qing", "Xin Yao", "Dong Huang", "Lin Zhang", "Zhuoran Ji"], "categories": ["cs.SE", "cs.CR"], "id": "2606.01317", "title": "SABER: Benchmarking Operational Safety of LLM Coding Agents in Stateful Project Workspaces", "type": "arxiv", "url": "https://arxiv.org/abs/2606.01317"}, "sourceUpdatedAt": "2026-05-31T16:06:02Z", "topics": ["Agents", "Safety"], "venueAttempts": [], "whyItMatters": "Existing safety benchmarks only check prompt refusal, missing the impact of action sequences on workspaces. SABER fills this gap by measuring environment-aware operational safety, offering a practical way to compare models on their ability to avoid harmful state changes in realistic coding tasks."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
