{
 "public_id": "AIR-2025-0164",
 "incident_id": "2025-01-lumia-aikatz-desktop-token-theft",
 "date_disclosed": "2025-11-01",
 "date_occurred": null,
 "title": "AIKatz: harvesting authentication tokens from ChatGPT, Claude and Copilot desktop apps to hijack LLM sessions",
 "class": "responsible_disclosure",
 "attack_surface": "enterprise_assistant",
 "initial_vector": "local_host_compromise",
 "agent_capability_abused": [
  "session_impersonation",
  "conversation_history_access",
  "context_poisoning"
 ],
 "target_product": "Claude, ChatGPT and Microsoft 365 Copilot desktop applications",
 "target_vendor": "Anthropic, OpenAI, Microsoft",
 "autonomy_level": "assistant",
 "impact_realized": false,
 "impact_type": [
  "session_hijack",
  "data_exfiltration",
  "integrity_violation"
 ],
 "blast_radius": "any user of the Electron/WebView-based LLM desktop clients on a compromised endpoint",
 "cve": [],
 "taxonomy": {
  "owasp_asi": [
   "ASI06",
   "ASI03"
  ],
  "mitre_atlas": [
   "AML.CS0036"
  ],
  "labels": [
   "post_breach",
   "token_theft",
   "chromium_client",
   "context_poisoning"
  ]
 },
 "defense_present": "OS-level credential protection",
 "defense_bypassed": true,
 "fix_shipped": "none - all three vendors closed the reports as informational / out of scope",
 "primary_sources": [
  "https://www.lumia.security/blog/aikatz",
  "https://raw.githubusercontent.com/mitre-atlas/atlas-data/main/dist/v6/ATLAS-2026.07.yaml"
 ],
 "verified_quote": "The attack allows, through credential harvesting, for the attacker to access and manipulate a victim’s LLM conversations after infecting their machine (post-breach exploitation).",
 "confidence": "high",
 "notes": "Lumia Security. Anthropic (HackerOne), MSRC and OpenAI (BugCrowd) all declined to treat it as a security boundary violation. ATLAS dates the case study to 2025 with year granularity; date_disclosed set to the ATLAS November 2025 publication month.",
 "risk_category": "security_fraud",
 "risk_secondary": [
  "privacy"
 ],
 "risk_subcategory": "Credential & Secret Theft",
 "risk_sub_secondary": [
  "Unauthorized Access & Account Takeover"
 ],
 "ai_entity": "generative",
 "agency": "ai_as_target",
 "agency_debatable": false,
 "agency_reason": "local host compromise harvesting tokens from LLM desktop apps",
 "vector_family": "other",
 "vector_detail": "local_host_compromise",
 "guardrail_kind": [
  "unknown"
 ],
 "guardrail_outcome": "silent",
 "xrefs": {
  "aiid": [],
  "atlas": [
   "AML.CS0036"
  ],
  "cve": [],
  "ghsa": []
 },
 "primary_asi": "ASI06",
 "agent_framework": "m365_copilot",
 "model_provider": "multiple",
 "model_name": "",
 "detection_mechanism": "unknown",
 "tool_access": "unknown",
 "financial_loss_band": "unknown",
 "eai_category": "credential_theft",
 "eai_domain": "access_control_and_permissions",
 "eai_secondary": [
  "data_exfiltration_channels",
  "confused_deputy",
  "data_and_memory_poisoning"
 ],
 "eai_reason": "subcategory: Credential & Secret Theft",
 "asi_curated": [],
 "asi_scope": "in_scope",
 "asi_provenance": "rule",
 "asi_reason": "ASI06 (signal=memory); ASI03 (eai_category=credential_theft)",
 "submitted_by": "EAI Internal"
}