{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.02018", "kind": "paper", "url": "https://arxiv.org/abs/2608.02018"}], "authors": ["Jia-Chen Zhang", "Ze-Yu Zhang", "Kai-Wei Zhang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.02018", "source_url": "https://arxiv.org/abs/2608.02018"}, "categories": ["Software & AI Compute", "Vision & 3D", "cs.CV"], "description": {"en": "II-Bench evaluates computer-use agents against low-harm adversarial tasks across three platforms, with 444 examples and a testing framework."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.02018", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.02018", "title": "II-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["East China Normal University", "Shanghai Jiao Tong University", "Shanghai Artificial Intelligence Laboratory"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2608.02018", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2608.02018", "source_url": "https://openalex.org/W7172437520"}, "key": "claire-radar:2608.02018", "modality": null, "name": "II-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_ii-bench_6645329e", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "505bee49cc9b936b24fa97340737a64a4c59f01ccd4d0f6453863d20ff6441f3", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2608.02018"}, "publisher": null, "released": "2026-08-03", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.02018", "source_url": "https://arxiv.org/abs/2608.02018"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-02018", "source": "claire_radar", "source_benchmark_id": "2608.02018", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Software & AI Compute"], "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": null, "hfPaperUrl": "https://huggingface.co/papers/2608.02018", "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": "7a0b6cccfe47f257d527480101c0136c923e632f86faa905e8e1e140fa873ce0", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "II-Bench evaluates computer-use agents against low-harm adversarial tasks across three platforms, with 444 examples and a testing framework.", "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": "To systematically investigate this blind spot, we present II-Bench, a collection of seemingly harmless adversarial tasks."}, "familyId": "bmf_e098118d24e6", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_ii-bench_6645329e", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Software & Cloud"], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2608.02018", "paper": "https://arxiv.org/abs/2608.02018", "pdf": "https://arxiv.org/pdf/2608.02018", "project": null, "report": "https://arxiv.org/abs/2608.02018"}, "metrics": [], "motivation": "Computer-use agents (CUAs), which empower large language models to autonomously operate operating systems and the web, are increasingly vulnerable to indirect prompt injection attacks.", "name": "II-Bench", "oneLine": "II-Bench evaluates computer-use agents against low-harm adversarial tasks across three platforms, with 444 examples and a testing framework.", "paperTitle": "Invisible Ink Threats: Adversarial Goals Behind Legitimate Tasks in Computer-Use Agents", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.02018", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {"90d": {"confidence": "Low", "coverage": 0.0, "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": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "rank": null, "score": null}, "method": "attention-ranking-v13", "rank": null, "score": null}}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-08-03", "source": {"authors": ["Jia-Chen Zhang", "Ze-Yu Zhang and Kai-Wei Zhang"], "categories": ["cs.CV"], "id": "2608.02018", "title": "Invisible Ink Threats: Adversarial Goals Behind Legitimate Tasks in Computer-Use Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2608.02018"}, "sourceUpdatedAt": "2026-08-06T02:33:08Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "Exposes security blind spots in human-in-the-loop defenses for computer-use agents."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
