{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.22643", "kind": "paper", "url": "https://arxiv.org/abs/2605.22643"}], "authors": ["Piercosma Bisconti", "Matteo Prandi", "Federico Pierucci", "Federico Sartore", "Enrico Panai", "Laura Caroli", "Yue Zhu", "Adam Leon Smith", "Luca Nannini", "Marcello Galisai", "Susanna Cifani", "Francesco Giarrusso", "Marcantonio Bracale Syrnikov", "Daniele Nardi"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.22643", "source_url": "https://arxiv.org/abs/2605.22643"}, "categories": ["General AI", "Safety", "Safety & Trustworthiness"], "description": {"en": "Boiling the Frog evaluates whether tool-using AI models in corporate settings are susceptible to incremental attacks through multi-turn scenarios with persistent workspaces. It includes a three-level operational risk taxonomy and scores attack success rate (ASR) on resulting artifact state."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.22643", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.22643", "title": "Boiling the Frog"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Sapienza University of Rome", "Scuola Superiore Sant'Anna", "Università Cattolica del Sacro Cuore", "Tongji University", "Clinical Research Consortium", "Vrije Universiteit Amsterdam"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2605.22643", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2605.22643", "source_url": "https://openalex.org/W7162112468"}, "key": "claire-radar:2605.22643", "modality": null, "name": "Boiling the Frog", "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_boiling-the-frog_8414ba51", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "078d051a0458c93da62c871117607dd23508729ae71639fb1eb6fb3dff876145", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.22643"}, "publisher": null, "released": "2026-05-21", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.22643", "source_url": "https://arxiv.org/abs/2605.22643"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-22643", "source": "claire_radar", "source_benchmark_id": "2605.22643", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Mixed", "applicationDomains": ["General AI"], "area": "Safety & Trustworthiness", "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/2605.22643", "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": "Aggregate Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "fcdac064900bc9e92b7202a5484c1389a775db9ac286461cf3194aed80de665c", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Boiling the Frog evaluates whether tool-using AI models in corporate settings are susceptible to incremental attacks through multi-turn scenarios with persistent workspaces. It includes a three-level operational risk taxonomy and scores attack success rate (ASR) on resulting artifact state.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence"], "snippet": "We introduce Boiling the Frog, a benchmark that evaluates whether tool-using AI models deployed in corporate and office settings are susceptible to incremental attacks."}, "familyId": "bmf_f1d48d3c8c74", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_boiling-the-frog_8414ba51", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2605.22643", "paper": "https://arxiv.org/abs/2605.22643", "pdf": "https://arxiv.org/pdf/2605.22643", "project": null, "report": "https://arxiv.org/abs/2605.22643"}, "metrics": [], "motivation": "Background.", "name": "Boiling the Frog", "oneLine": "Boiling the Frog evaluates whether tool-using AI models in corporate settings are susceptible to incremental attacks through multi-turn scenarios with persistent workspaces. It includes a three-level operational risk taxonomy and scores attack success rate (ASR) on resulting artifact state.", "paperTitle": "Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.22643", "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": "2026-05-21", "source": {"authors": ["Piercosma Bisconti", "Matteo Prandi", "Federico Pierucci", "Federico Sartore", "Enrico Panai", "Laura Caroli", "Yue Zhu", "Adam Leon Smith", "Luca Nannini", "Marcello Galisai", "Susanna Cifani", "Francesco Giarrusso", "Marcantonio Bracale Syrnikov", "Daniele Nardi"], "categories": ["cs.CL"], "id": "2605.22643", "title": "Boiling the Frog: A Multi-Turn Benchmark for Agentic Safety", "type": "arxiv", "url": "https://arxiv.org/abs/2605.22643"}, "sourceUpdatedAt": "2026-05-22T14:53:30Z", "topics": ["Safety"], "venueAttempts": [], "whyItMatters": "Addresses the gap in safety evaluation for agents acting in environments, focusing on incremental manipulation rather than single-turn textual outputs."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
