{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.25592", "kind": "paper", "url": "https://arxiv.org/abs/2606.25592"}, {"id": "hf:csu-jpg/vva-bench", "kind": "dataset", "url": "https://huggingface.co/datasets/CSU-JPG/VVA-Bench"}], "categories": ["General AI", "Safety", "Safety & Trustworthiness"], "description": {"en": "A safety benchmark for image-to-video systems exposed to categorized attacks encoded through visual prompts."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.25592", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.25592", "title": "VVA-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.25592", "modality": null, "name": "VVA-Bench", "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_vva-bench_87789f6a", "origin_source": "arxiv", "recognition_confidence": "0.55", "record_sha256": "dc3e076e9ff4c6c6e965a108154acab0bb23678ab2977b4078aa6567a4b91c0d", "relation": "evaluates_only", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T20:22:34.615233Z", "source_url": "https://arxiv.org/abs/2606.25592"}, "publisher": null, "released": "2026-06-24", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.25592", "source_url": "https://arxiv.org/abs/2606.25592"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-25592", "source": "claire_radar", "source_benchmark_id": "2606.25592", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Safety & Trustworthiness", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Video-generation safety, visual prompt attacks, and reported attack success rates as high as 100% create a strong near-term research hook.", "score": 73}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T20:22:34.615233Z", "inputHash": "cd79e1f240a8580e0ac190f64e567ea91689d5f7b25acf61cdd25dfe33ecb7ff", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "we present VVA-Bench, the first systematic benchmark", "canonicalNameSource": "abstract", "decisionReason": "The source formally names the benchmark, specifies the evaluated attack setting, reports a concrete metric, and supplies a public dataset page.", "evaluationEvidence": {"reuse": "Dataset Page: https://huggingface.co/datasets/CSU-JPG/VVA-Bench", "scoring": "Attack Success Rates (ASR) reaching 100.0\\% on Wan 2.7 and 74.8\\% on Veo 3.1", "task": "evaluating video generation safety under categorized vision-centric prompt attacks"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T20:22:34.615233Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A safety benchmark for image-to-video systems exposed to categorized attacks encoded through visual prompts.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "public artifact URL", "named entity is a model, method, framework, or dataset used in evaluation"], "snippet": "To bridge this gap, we present VVA-Bench, the first systematic benchmark for evaluating video generation safety under categorized vision-centric prompt attacks."}, "familyId": "bmf_f64c7ad6d93c", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_vva-bench_87789f6a", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": "https://huggingface.co/datasets/CSU-JPG/VVA-Bench", "paper": "https://arxiv.org/abs/2606.25592", "pdf": "https://arxiv.org/pdf/2606.25592", "project": null, "report": "https://arxiv.org/abs/2606.25592"}, "metrics": [], "motivation": "Recent advancements in Image-to-Video (I2V) generation have transformed input images from simple appearance references into interactive control interfaces where visual cues such as arrows, sketches, and emojis orchestrate complex video dynamics with unprecedented controllability.", "name": "VVA-Bench", "oneLine": "A safety benchmark for image-to-video systems exposed to categorized attacks encoded through visual prompts.", "paperTitle": "VPA-Guard: Defending and Benchmarking Image-to-Video Generation Against Visual Prompt Attacks", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.25592", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "publishers": [{"name": "CSU-JPG", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://huggingface.co/datasets/CSU-JPG/VVA-Bench"}], "readiness": "Inspectable", "recognitionConfidence": 0.55, "relation": "evaluates_only", "releasedAt": "2026-06-24", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Yining Sun", "Haoyu Kang", "Jiajun Wu", "Heng Zhang", "Danyang Zhang", "Zhenjun Zhao", "Haochen Han", "Fangming Liu", "Wai Kin Victor Chan", "Alex Jinpeng Wang"], "categories": ["cs.CV"], "id": "2606.25592", "title": "VPA-Guard: Defending and Benchmarking Image-to-Video Generation Against Visual Prompt Attacks", "type": "arxiv", "url": "https://arxiv.org/abs/2606.25592"}, "sourceUpdatedAt": "2026-06-24T09:00:08Z", "topics": ["Safety"], "venueAttempts": [], "whyItMatters": "It evaluates a safety surface not covered by predominantly text-based or content-only image jailbreak benchmarks and reports attack success rates."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
