{
 "public_id": "AIR-2026-0272",
 "incident_id": "2026-10-aws-agentcore-agentcorruption",
 "date_disclosed": "2026-10-08",
 "date_occurred": null,
 "title": "AgentCorruption: AgentCore metadata credentials enable regional lateral access and persistent memory-driven conversation exfiltration",
 "class": "responsible_disclosure",
 "attack_surface": "agent_framework",
 "initial_vector": "direct_prompt_induced_metadata_request",
 "agent_capability_abused": [
  "http_tool",
  "shell_tool",
  "execution_role_credential_access",
  "agent_invocation",
  "container_image_access",
  "conversation_memory_read_write",
  "memory_retrieval",
  "outbound_conversation_transmission"
 ],
 "target_product": "Amazon Bedrock AgentCore Runtime and Memory; demonstrated with a Strands SDK agent",
 "target_vendor": "Amazon Web Services",
 "autonomy_level": "semi_autonomous",
 "impact_realized": false,
 "impact_type": [
  "credential_theft",
  "unauthorized_access",
  "information_disclosure",
  "memory_poisoning",
  "agent_hijack",
  "data_exfiltration"
 ],
 "blast_radius": "All AgentCore agents in the same AWS account and region (demonstrated in Zenity's own account)",
 "cve": [],
 "taxonomy": {
  "mitre_atlas": [],
  "labels": [
   "AgentCorruption",
   "metadata_service",
   "ssrf",
   "execution_role",
   "overprivileged_iam",
   "lateral_movement",
   "memory_poisoning",
   "persistent_hijacking",
   "vendor_dispute"
  ],
  "owasp_asi": [
   "ASI01",
   "ASI06",
   "ASI03"
  ]
 },
 "defense_present": "Firecracker microVM isolation; execution-role IAM permissions",
 "defense_bypassed": true,
 "fix_shipped": "New agents use IMDSv2 only from 14 Feb 2026; AWS closed the report as informative",
 "primary_sources": [
  "https://labs.zenity.io/post/agentcorruption-initial-imds-access",
  "https://labs.zenity.io/post/agentcorruption-weaponizing-agent-memory-for-persistent-hijacking",
  "https://zenity.io/press-release/zenity-labs-discloses-agentcorruption-a-chain-of-aws-agentcore-flaws"
 ],
 "verified_quote": "It worked. The microVM did not restrict traffic to the metadata service, and the agent happily relayed the response back to us.",
 "confidence": "high",
 "notes": "Zenity Labs (Lana Salameh, Dmitry Lozovoy), five-part series published 8 Oct 2026; reported to AWS in December 2025 (the post gives both 17 and 25 Dec). A plain prompt made a Strands agent with an HTTP tool fetch the metadata endpoint and send back its execution-role credentials, which worked from outside AgentCore. The default role's wildcard permissions then reached every agent in the same account and region: container images, conversations, memory and secrets. Poisoned memory gave persistence. Everything was demonstrated in Zenity's own account; no customer victims. AWS says the behavior is documented and expected and recommends least-privilege roles.",
 "risk_category": "security_fraud",
 "risk_secondary": [
  "privacy"
 ],
 "risk_subcategory": "Server-Side Request Forgery via LLM",
 "risk_sub_secondary": [
  "Credential & Secret Theft",
  "Unauthorized Access & Account Takeover",
  "Over-permissioned execution & lateral movement"
 ],
 "ai_entity": "generative",
 "agency": "agent_acted",
 "agency_debatable": false,
 "agency_reason": "the demonstrated agent queries metadata credentials and later retrieves attacker instructions and transmits conversations through tools",
 "vector_family": "direct_prompt_injection",
 "vector_detail": "direct_prompt_induced_metadata_request",
 "guardrail_kind": [
  "sandbox_isolation",
  "permission_scope"
 ],
 "guardrail_outcome": "silent",
 "xrefs": {
  "aiid": [],
  "atlas": [],
  "cve": [],
  "ghsa": []
 },
 "primary_asi": "ASI01",
 "agent_framework": "unknown",
 "model_provider": "unknown",
 "model_name": "",
 "detection_mechanism": "external_researcher",
 "tool_access": "unknown",
 "financial_loss_band": "unknown",
 "eai_category": "confused_deputy",
 "eai_domain": "access_control_and_permissions",
 "eai_secondary": [
  "data_and_memory_poisoning",
  "credential_theft",
  "privilege_escalation",
  "data_exfiltration_channels",
  "sensitive_data_exposure"
 ],
 "eai_reason": "subcategory: Server-Side Request Forgery via LLM",
 "asi_curated": [],
 "asi_scope": "in_scope",
 "asi_provenance": "rule",
 "asi_reason": "ASI01 (vector_family=direct_prompt_injection); ASI06 (signal=memory); ASI03 (eai_category=confused_deputy)",
 "submitted_by": "EAI Internal"
}