{
 "id": "AIR-2023-0002",
 "slug": "bing-chat-data-exfiltration-via-indirect-prompt-injection",
 "title": "Bing Chat Data Exfiltration via Indirect Prompt Injection",
 "summary": "Researchers showed that visiting malicious websites while using Bing Chat allowed indirect prompt injections to covertly exfiltrate personal user data. The attack turned the chatbot into a social engineering tool without user consent.",
 "type": "attack_on_ai",
 "categories": [
  "prompt_injection",
  "agent_hijack",
  "data_leak"
 ],
 "severity": 1,
 "targets": [
  "Microsoft"
 ],
 "products": [
  "Bing Chat"
 ],
 "incident_date": "2023",
 "incident_date_precision": "year",
 "first_reported": "2023-10-30",
 "sources": [
  {
   "url": "https://atlas.mitre.org/studies/AML.CS0020",
   "publisher": "MITRE ATLAS",
   "title": "Indirect Prompt Injection Threats: Bing Chat Data Pirate",
   "published_at": "2023-10-30"
  },
  {
   "url": "https://greshake.github.io/",
   "publisher": "greshake.github.io",
   "title": "Indirect Prompt Injection Threats: Bing Chat Data Pirate",
   "published_at": "2023-10-30"
  }
 ],
 "source_count": 2,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_atlas",
 "tags": [
  "indirect_prompt_injection",
  "data_exfiltration"
 ],
 "created_at": "2026-10-18T06:55:08.239Z",
 "updated_at": "2026-10-18T06:55:08.239Z"
}