{
 "public_id": "AIR-2026-0270",
 "incident_id": "2026-10-databricks-genie-malicious-skills-exfiltration",
 "date_disclosed": "2026-10-05",
 "date_occurred": null,
 "title": "Databricks Genie: malicious Skills bridge compute and display controls to exfiltrate datasets and render phishing overlays",
 "class": "responsible_disclosure",
 "attack_surface": "skill_plugin",
 "initial_vector": "malicious_uploaded_skill",
 "agent_capability_abused": [
  "skill_code_execution",
  "tenant_dataset_query",
  "html_display_generation",
  "browser_network_requests"
 ],
 "target_product": "Databricks Genie Code",
 "target_vendor": "Databricks",
 "autonomy_level": "semi_autonomous",
 "impact_realized": false,
 "impact_type": [
  "data_exfiltration",
  "phishing"
 ],
 "blast_radius": "Genie Code users who load a malicious Skill from their personal workspace",
 "cve": [],
 "taxonomy": {
  "mitre_atlas": [],
  "labels": [
   "malicious_skill",
   "agent_in_the_loop",
   "display_exfiltration",
   "split_environment_controls",
   "personal_workspace",
   "vendor_dispute"
  ],
  "owasp_asi": [
   "ASI04",
   "ASI05"
  ]
 },
 "defense_present": "Guardrail agent (auto-approve mode), coding-environment egress controls, display sandbox",
 "defense_bypassed": true,
 "fix_shipped": null,
 "primary_sources": [
  "https://www.promptarmor.com/resources/phishing-in-databricks-genie"
 ],
 "verified_quote": "The controls to prevent exfiltration were technically upheld, but exfiltration succeeded nonetheless.",
 "confidence": "high",
 "notes": "PromptArmor; reported to Databricks on 16 Aug 2026 and published 5 Oct. A malicious Skill loaded from the user's personal workspace (not the governed catalog) passed the guardrail agent, queried tenant data and built it into an HTML display. Viewing the results showed a credential-phishing overlay and sent the data out from the user's browser, which the coding environment's egress controls do not cover. Researcher demonstration; no victims reported. Databricks says malicious uploaded Skills are the user's responsibility and that auto-approve is not a security boundary. No fix.",
 "risk_category": "privacy",
 "risk_secondary": [],
 "risk_subcategory": "Data Exfiltration via Context",
 "risk_sub_secondary": [],
 "ai_entity": "generative",
 "agency": "agent_acted",
 "agency_debatable": true,
 "agency_reason": "Genie executes the skill, queries tenant data and constructs a malicious display; user viewing triggers the final browser transmission",
 "vector_family": "malicious_component",
 "vector_detail": "malicious_uploaded_skill",
 "guardrail_kind": [
  "human_approval",
  "sandbox_isolation",
  "permission_scope"
 ],
 "guardrail_outcome": "silent",
 "xrefs": {
  "aiid": [],
  "atlas": [],
  "cve": [],
  "ghsa": []
 },
 "primary_asi": "ASI04",
 "agent_framework": "unknown",
 "model_provider": "unknown",
 "model_name": "",
 "detection_mechanism": "external_researcher",
 "tool_access": "unknown",
 "financial_loss_band": "unknown",
 "eai_category": "data_exfiltration_channels",
 "eai_domain": "privacy",
 "eai_secondary": [
  "human_manipulation",
  "supply_chain_vulnerabilities"
 ],
 "eai_reason": "subcategory: Data Exfiltration via Context",
 "asi_curated": [],
 "asi_scope": "in_scope",
 "asi_provenance": "rule",
 "asi_reason": "ASI04 (vector_family=malicious_component); ASI05 (vector=code_exec)",
 "submitted_by": "EAI Internal"
}