{"record": {"aliases": ["PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say"], "artifacts": [{"id": "arxiv:2606.00152", "kind": "paper", "url": "https://arxiv.org/abs/2606.00152"}, {"id": "gh:xuan269/privacypeek-resource", "kind": "repo", "url": "https://github.com/Xuan269/PrivacyPeek-Resource"}], "authors": ["Mingxuan Zhang", "Jiahui Han", "Dadi Guo", "Songze Li", "Guanchu Wang", "Na Zou", "Dongrui Liu", "Xia Hu"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.00152", "source_url": "https://arxiv.org/abs/2606.00152"}, "categories": ["General AI", "Language & Knowledge", "cs.CR"], "description": {"en": "A 1,182-case benchmark across 7 acquisition behaviours and 16 domains that audits whether tool-using LLM agents acquire sensitive data beyond task scope."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.00152", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.00152", "title": "PrivacyPeek"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Shanghai Artificial Intelligence Laboratory", "Southeast University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.00152", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.00152", "source_url": "https://openalex.org/W7163178005"}, "key": "claire-radar:2606.00152", "modality": null, "name": "PrivacyPeek", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_privacypeek-auditing-what-llm-based-agents_6fde790d", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "3d894ea8e98ceb2e96a561c02e9f6336c7552886d0f53998029d1f922062ae08", "relation": "introduces", "review_model": "claude-code-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T16:50:07.444368Z", "source_url": "https://arxiv.org/abs/2606.00152"}, "publisher": null, "released": "2026-05-29", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.00152", "source_url": "https://arxiv.org/abs/2606.00152"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-00152", "source": "claire_radar", "source_benchmark_id": "2606.00152", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "Agent privacy and security is an active topic; the acquisition-stage framing is new.", "score": 45}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T16:50:07.444368Z", "inputHash": "48e792f2266c8828995d03207554a41df8ab53a4e1388681e841aac9e94b075b", "model": "claude-code-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say", "canonicalNameSource": "paper_title", "decisionReason": "Named benchmark with defined Acquisition Inspection and Probe Elicitation scoring procedures, plus a public dataset and code.", "evaluationEvidence": {"reuse": "Our dataset and code are available at https://github.com/Xuan269/PrivacyPeek-Resource.", "scoring": "examines the agent's tool-call trajectory, both the tools it invokes and the data it receives", "task": "a benchmark for evaluating acquisition-stage privacy leakage of LLM-based agents"}, "model": "claude-code-default", "reviewedAt": "2026-09-24T16:50:07.444368Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A 1,182-case benchmark across 7 acquisition behaviours and 16 domains that audits whether tool-using LLM agents acquire sensitive data beyond task scope.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "To assess its prevalence, we introduce \\emph{PrivacyPeek}, a benchmark for evaluating acquisition-stage privacy leakage of LLM-based agents, with $1{,}182$ cases across $7$ acquisition behaviours and $16$ application domains."}, "familyId": "bmf_ee170dacaa41", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_privacypeek-auditing-what-llm-based-agents_6fde790d", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/Xuan269/PrivacyPeek-Resource", "data": null, "paper": "https://arxiv.org/abs/2606.00152", "pdf": "https://arxiv.org/pdf/2606.00152", "project": null, "report": "https://arxiv.org/abs/2606.00152"}, "metrics": [], "motivation": "LLM-based agents are rapidly advancing, autonomously invoking external tools to complete multi-step tasks for users.", "name": "PrivacyPeek", "oneLine": "A 1,182-case benchmark across 7 acquisition behaviours and 16 domains that audits whether tool-using LLM agents acquire sensitive data beyond task scope.", "paperTitle": "PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.00152", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-05-29", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Mingxuan Zhang", "Jiahui Han", "Dadi Guo", "Songze Li", "Guanchu Wang", "Na Zou", "Dongrui Liu", "Xia Hu"], "categories": ["cs.CR", "cs.AI"], "id": "2606.00152", "title": "PrivacyPeek: Auditing What LLM-Based Agents Acquire, Not Just What They Say", "type": "arxiv", "url": "https://arxiv.org/abs/2606.00152"}, "sourceUpdatedAt": "2026-08-06T13:29:02Z", "topics": ["cs.CR"], "venueAttempts": [], "whyItMatters": "It moves privacy evaluation of agents from disclosure to acquisition, measuring over-acquisition in tool trajectories and how easily a follow-up probe can elicit that data."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
