{
 "public_id": "AIR-2024-0049",
 "incident_id": "2024-12-xai-grok-security-probllms",
 "date_disclosed": "2024-12-16",
 "date_occurred": null,
 "title": "Security ProbLLMs in xAI's Grok: prompt injection, ASCII smuggling and image-render data exfiltration",
 "class": "responsible_disclosure",
 "attack_surface": "consumer_chatbot",
 "initial_vector": "indirect_prompt_injection_social_post",
 "agent_capability_abused": [
  "retrieval",
  "markdown_image_rendering",
  "document_analysis"
 ],
 "target_product": "Grok (web and iOS app)",
 "target_vendor": "xAI",
 "autonomy_level": "assistant",
 "impact_realized": false,
 "impact_type": [
  "data_exfiltration",
  "pii_disclosure",
  "disinformation"
 ],
 "blast_radius": "Grok users; targeted conditional payloads possible per-user",
 "cve": [],
 "taxonomy": {
  "owasp_asi": [
   "ASI01",
   "ASI02"
  ],
  "mitre_atlas": [],
  "labels": [
   "ascii_smuggling",
   "image_markdown_exfiltration",
   "conditional_injection",
   "no_csp"
  ]
 },
 "defense_present": "none - third-party image URLs rendered without sandboxing",
 "defense_bypassed": true,
 "fix_shipped": "not fixed - xAI closed all tickets as Informational; a partial change later prevented username inclusion; the January 2025 Grok iOS app had the same issue",
 "primary_sources": [
  "https://embracethered.com/blog/posts/2024/security-probllms-in-xai-grok/"
 ],
 "verified_quote": "All issues mentioned in this post were responsibly disclosed to xAI.",
 "confidence": "high",
 "notes": "Johann Rehberger / Embrace The Red. Leaked user chat history, IP addresses and usernames via rendered images.",
 "risk_category": "privacy",
 "risk_secondary": [
  "product_misuse"
 ],
 "risk_subcategory": "PII Leakage",
 "risk_sub_secondary": [
  "Data Exfiltration via Context"
 ],
 "ai_entity": "generative",
 "agency": "agent_acted",
 "agency_debatable": false,
 "agency_reason": "Grok rendered images exfiltrating chat history and IPs",
 "vector_family": "indirect_prompt_injection",
 "vector_detail": "indirect_prompt_injection_social_post",
 "guardrail_kind": [
  "none"
 ],
 "guardrail_outcome": "not_applicable",
 "xrefs": {
  "aiid": [],
  "atlas": [],
  "cve": [],
  "ghsa": []
 },
 "primary_asi": "ASI01",
 "agent_framework": "unknown",
 "model_provider": "xai",
 "model_name": "",
 "detection_mechanism": "external_researcher",
 "tool_access": "unknown",
 "financial_loss_band": "unknown",
 "eai_category": "sensitive_data_exposure",
 "eai_domain": "privacy",
 "eai_secondary": [
  "data_exfiltration_channels",
  "confused_deputy"
 ],
 "eai_reason": "subcategory: PII Leakage",
 "asi_curated": [
  "ASI01"
 ],
 "asi_scope": "in_scope",
 "asi_provenance": "curated_plus_rule",
 "asi_reason": "curated: ASI01 | added ASI02 (impact=tool_misuse)",
 "submitted_by": "EAI Internal"
}