{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.12911", "kind": "paper", "url": "https://arxiv.org/abs/2608.12911"}], "authors": ["Beining Xu", "Hairui Wang", "Jiaxin Wang", "Changsheng Chen", "Anirban Chakraborty"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.12911", "source_url": "https://arxiv.org/abs/2608.12911"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "DocPrivacyBench evaluates susceptibility of document understanding MLLMs to relational privacy leakage when visual evidence is absent or minimal, using KIE tasks on identity documents."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.12911", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.12911", "title": "DocPrivacyBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Shenzhen MSU-BIT University", "Indian Institute of Science Bangalore"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2608.12911", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2608.12911", "source_url": "https://openalex.org/W7203455449"}, "key": "claire-radar:2608.12911", "modality": null, "name": "DocPrivacyBench", "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_docprivacybench_1e095c41", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "424c7e02e98d1b2576f06d093deddd71f16b25ab847b28c89233f5ed16d70450", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2608.12911"}, "publisher": null, "released": "2026-08-13", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.12911", "source_url": "https://arxiv.org/abs/2608.12911"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-12911", "source": "claire_radar", "source_benchmark_id": "2608.12911", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "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": 0, "hfPaperUrl": "https://huggingface.co/papers/2608.12911", "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", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "stale"}, "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": "b077a7d75b0671f82623b40664d07463f3e8457ba5233ff336ba1821edca71df", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "DocPrivacyBench evaluates susceptibility of document understanding MLLMs to relational privacy leakage when visual evidence is absent or minimal, using KIE tasks on identity documents.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "It suppresses the leakage of high-risk field pairs while preserving KIE performance.Second, we introduce DocPrivacyBench, a novel benchmark to systematically evaluate a model's susceptibility to privacy leakage under conditions of absent or minimal visual evidence.Third, we evaluate three MLLMs and six unlearning methods using this benchmark, assessing both post-unlearning leakage suppression and utility preservation.Our results demonstrate that existing MLLMs consistently exhibit privacy leakag"}, "familyId": "bmf_db22e09b280e", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_docprivacybench_1e095c41", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2608.12911", "paper": "https://arxiv.org/abs/2608.12911", "pdf": "https://arxiv.org/pdf/2608.12911", "project": null, "report": "https://arxiv.org/abs/2608.12911"}, "metrics": [], "motivation": "While the privacy risks of multimodal large language models (MLLMs) have drawn significant attention, the unique vulnerabilities of domain-specific MLLMs remain largely underexplored.", "name": "DocPrivacyBench", "oneLine": "DocPrivacyBench evaluates susceptibility of document understanding MLLMs to relational privacy leakage when visual evidence is absent or minimal, using KIE tasks on identity documents.", "paperTitle": "Beyond Visual Evidence: Revealing and Mitigating Relational Privacy Leakage in Document MLLMs", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.12911", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {"90d": {"confidence": "Low", "coverage": 0.15, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": null, "value": null}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.0, "value": 0}}, "confidence": "Low", "coverage": 0.15, "rank": null, "score": 43}, "method": "attention-ranking-v13", "rank": null, "score": 43}}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-08-13", "source": {"authors": ["Beining Xu", "Hairui Wang", "Jiaxin Wang", "Changsheng Chen", "and Anirban Chakraborty"], "categories": ["cs.CV", "cs.CR", "cs.MM"], "id": "2608.12911", "title": "Beyond Visual Evidence: Revealing and Mitigating Relational Privacy Leakage in Document MLLMs", "type": "arxiv", "url": "https://arxiv.org/abs/2608.12911"}, "sourceUpdatedAt": "2026-08-13T07:53:53Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "Addresses underexplored privacy vulnerabilities in document MLLMs, providing a framework to assess and mitigate leakage of correlated sensitive fields, which is crucial for trustworthy deployment in document processing."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
