{
 "public_id": "AIR-2026-0030",
 "incident_id": "2026-02-openclaw-prompt-injection-c2",
 "date_disclosed": "2026-02-03",
 "date_occurred": null,
 "title": "Turning OpenClaw into a persistent command-and-control implant via indirect prompt injection",
 "class": "research_demo",
 "attack_surface": "agent_framework",
 "initial_vector": "indirect_prompt_injection_web_page",
 "agent_capability_abused": [
  "web_browsing",
  "shell_execution",
  "system_prompt_persistence"
 ],
 "target_product": "OpenClaw",
 "target_vendor": "OpenClaw project",
 "autonomy_level": "fully_autonomous",
 "impact_realized": false,
 "impact_type": [
  "persistence",
  "code_execution",
  "remote_control"
 ],
 "blast_radius": "OpenClaw installs that browse untrusted web content",
 "cve": [],
 "taxonomy": {
  "owasp_asi": [
   "ASI01",
   "ASI05",
   "ASI03"
  ],
  "mitre_atlas": [
   "AML.CS0051"
  ],
  "labels": [
   "agent_as_implant",
   "persistence_via_memory_file",
   "reasoning_spoofing",
   "HEARTBEAT.md"
  ]
 },
 "defense_present": "tool-approval prompts",
 "defense_bypassed": true,
 "fix_shipped": null,
 "primary_sources": [
  "https://www.hiddenlayer.com/research/exploring-the-security-risks-of-ai-assistants-like-openclaw",
  "https://raw.githubusercontent.com/mitre-atlas/atlas-data/main/dist/v6/ATLAS-2026.07.yaml"
 ],
 "verified_quote": "we’ll walk through an example attack using an indirect prompt injection embedded in a web page, which causes OpenClaw to install an attacker-controlled set of instructions in its HEARTBEAT.md file, causing the OpenClaw agent to silently wait for instructions from the attacker’s command and control server.",
 "confidence": "high",
 "notes": "HiddenLayer. Injection spoofs the model's <think> control sequences to make the hijack more reliable.",
 "risk_category": "security_fraud",
 "risk_secondary": [
  "privacy"
 ],
 "risk_subcategory": "Prompt Injection",
 "risk_sub_secondary": [
  "Remote Code Execution & Sandbox Escape",
  "Persistence & Covert C2"
 ],
 "ai_entity": "generative",
 "agency": "agent_acted",
 "agency_debatable": false,
 "agency_reason": "the injected agent persisted and executed C2 commands",
 "vector_family": "indirect_prompt_injection",
 "vector_detail": "indirect_prompt_injection_web_page",
 "guardrail_kind": [
  "human_approval"
 ],
 "guardrail_outcome": "silent",
 "xrefs": {
  "aiid": [],
  "atlas": [
   "AML.CS0051"
  ],
  "cve": [],
  "ghsa": []
 },
 "primary_asi": "ASI01",
 "agent_framework": "openclaw",
 "model_provider": "unknown",
 "model_name": "",
 "detection_mechanism": "external_researcher",
 "tool_access": "unknown",
 "financial_loss_band": "unknown",
 "eai_category": "privilege_escalation",
 "eai_domain": "access_control_and_permissions",
 "eai_secondary": [
  "confused_deputy"
 ],
 "eai_reason": "impact rule: privilege_escalation",
 "asi_curated": [
  "ASI01"
 ],
 "asi_scope": "in_scope",
 "asi_provenance": "curated_plus_rule",
 "asi_reason": "curated: ASI01 | added ASI05 (impact=code_exec), ASI03 (signal=identity)",
 "submitted_by": "EAI Internal"
}