{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreicgytics76prdfkbas26driucmdxfpgahwmvdshniv6la37uwjh5a",
    "uri": "at://did:plc:pgryn3ephfd2xgft23qokfzt/app.bsky.feed.post/3mng3co6x7eu2"
  },
  "path": "/t/genai-agentic-ai-security-incidents-7-725-real-world-research-incidents/176510#post_1",
  "publishedAt": "2026-06-03T20:02:02.000Z",
  "site": "https://discuss.huggingface.co",
  "tags": [
    "https://huggingface.co/datasets/emmanuelgjr/genai-incidents"
  ],
  "textContent": "**Just released: GenAI & Agentic AI Security Incidents**\n\nA dataset of **7,725 real-world and research incidents** covering:\n\nPrompt injection, jailbreaks, data exfiltration, deepfakes, agent hijacking, AI-enabled harms, and more.\n\nEach incident is mapped to:\n\n**OWASP LLM Top 10 2025**\n**OWASP Agentic AI Security Initiative Top 10**\n**NIST AI RMF**\n**MITRE ATLAS techniques and tactics**\n\n\n    from datasets import load_dataset\n\n    ds = load_dataset(\"emmanuelgjr/genai-incidents\", split=\"train\")\n\n    # Example: filter for prompt injection incidents\n    prompt_injection = ds.filter(\n        lambda r: \"LLM01\" in (r[\"owasp_llm\"] or [])\n    )\n\n\n\nEach entry includes a `quality_tier` field — `curated`, `reviewed`, or `auto` — so researchers and practitioners can filter by vetting level.\n\nLicensed under **CC-BY-4.0** and citable with a **DOI**.\n\nFeedback, issues, and PRs are very welcome.\n\nhttps://huggingface.co/datasets/emmanuelgjr/genai-incidents",
  "title": "GenAI & Agentic AI Security Incidents — 7,725 real-world & research incidents"
}