{
 "id": "AIR-2025-0059",
 "slug": "data-exfiltration-via-prompt-injections-in-copilot-studio-ag",
 "title": "Data Exfiltration via Prompt Injections in Copilot Studio Agent",
 "summary": "Researchers demonstrated how an organization's data could be exfiltrated through prompt injections targeting an AI customer service agent. The attack exploited tool access in a Copilot Studio implementation.",
 "type": "attack_on_ai",
 "categories": [
  "prompt_injection",
  "agent_hijack",
  "data_leak"
 ],
 "severity": 1,
 "targets": [
  "Microsoft",
  "enterprise_deployment"
 ],
 "products": [
  "Copilot Studio"
 ],
 "incident_date": "2025-06",
 "incident_date_precision": "month",
 "first_reported": "2025-11-07",
 "sources": [
  {
   "url": "https://atlas.mitre.org/studies/AML.CS0037",
   "publisher": "MITRE ATLAS",
   "title": "Data Exfiltration via Agent Tools in Copilot Studio",
   "published_at": "2025-11-07"
  },
  {
   "url": "https://labs.zenity.io/p/a-copilot-studio-story-discovery-phase-in-ai-agents-f917",
   "publisher": "Zenity Labs",
   "title": "AgentFlayer: Discovery Phase of AI Agents in Copilot Studio",
   "published_at": "2025-11-07"
  },
  {
   "url": "https://labs.zenity.io/p/a-copilot-studio-story-2-when-aijacking-leads-to-full-data-exfiltration-bc4a",
   "publisher": "Zenity Labs",
   "title": "AgentFlayer: When AIjacking Leads to Full Data Exfiltration in Copilot Studio",
   "published_at": "2025-11-07"
  }
 ],
 "source_count": 3,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_atlas",
 "tags": [
  "prompt-injection",
  "copilot-studio",
  "data-exfiltration",
  "agent-hijack"
 ],
 "created_at": "2026-10-18T06:55:08.239Z",
 "updated_at": "2026-10-18T06:55:08.239Z"
}