{
  "id": "agentcorruption-bedrock-2026-10",
  "name": "AgentCorruption (AWS Bedrock AgentCore IMDS)",
  "date_reported": "2026-10-08",
  "researcher": "Tamir Ishay Sharbat, Lana Salameh, Dmitry Lozovoy",
  "org": "Zenity Labs",
  "sources": [
    {
      "url": "https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model"
    }
  ],
  "magicsword_coverage": [],
  "borderline": false,
  "borderline_note": null,
  "ai_role": "ai_agent_or_mcp_as_target",
  "ai_role_secondary": null,
  "ai_role_detail": null,
  "technique_summary": "Chat prompt to public AgentCore agent abused tools that could HTTP/shell to IMDS 169.254.169.254, stealing execution-role credentials. Default role was region-wide across AgentCore: invoke other agents, read conversations/memories, pull images, Secrets Manager / AgentCore Identity, plant persistent memory hijacks. AWS tightened IMDSv2 defaults and narrowed default role.",
  "initial_access": "Prompt injection / SSRF via agent tools against exposed AgentCore agent",
  "ai_tooling_and_models": [
    "AWS Bedrock AgentCore"
  ],
  "other_tooling": [
    "IMDS / IMDSv2",
    "AWS Secrets Manager",
    "ECR"
  ],
  "victims_and_sectors": "Organizations deploying AgentCore agents (research disclosure; patched)",
  "regions": [
    "unknown"
  ],
  "attribution_or_cluster": "Vulnerability research (Zenity at SecTor)",
  "scale_and_metrics": "Single prompt to one agent → account/region AgentCore blast radius (pre-fix)",
  "prevention_angle": "Least-privilege agent IAM; IMDSv2; agent's role is its blast radius",
  "related_magicsword_projects": [],
  "source_notes": null,
  "last_reviewed": "2026-10-20",
  "attack_flow": {
    "nodes": [
      {
        "id": "step-1",
        "title": "Public agent receives a prompt",
        "description": "An exposed AWS AgentCore agent accepts attacker input.",
        "kind": "access",
        "source_urls": [
          "https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model"
        ]
      },
      {
        "id": "step-2",
        "title": "Agent tools reach metadata",
        "description": "Prompt injection abuses HTTP or shell tools to query the instance metadata service.",
        "kind": "ai",
        "source_urls": [
          "https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model"
        ]
      },
      {
        "id": "step-3",
        "title": "Execution credentials are exposed",
        "description": "The response discloses the agent execution-role credentials.",
        "kind": "execution",
        "source_urls": [
          "https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model"
        ]
      },
      {
        "id": "step-4",
        "title": "Broader agent access",
        "description": "The role can reach other agents, memories and connected services; persistent memory hijacks are described.",
        "kind": "impact",
        "source_urls": [
          "https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model"
        ]
      }
    ],
    "edges": [
      {
        "source": "step-1",
        "target": "step-2"
      },
      {
        "source": "step-2",
        "target": "step-3"
      },
      {
        "source": "step-3",
        "target": "step-4"
      }
    ]
  },
  "atlas_techniques": [
    {
      "id": "AML.T0051.000",
      "rationale": "A prompt sent directly to the public agent redirects its tool use.",
      "source_urls": [
        "https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model"
      ]
    },
    {
      "id": "AML.T0055",
      "rationale": "The metadata response exposes execution-role credentials.",
      "source_urls": [
        "https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model"
      ]
    },
    {
      "id": "AML.T0080.000",
      "rationale": "The source describes persistent memory hijacks.",
      "source_urls": [
        "https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model"
      ]
    }
  ],
  "category": "attacks-on-ai",
  "attribution": {
    "actor_name": "Vulnerability research (Zenity at SecTor)"
  },
  "ai_involvement_confidence": "confirmed",
  "detections": [
    {
      "name": "Review attack-specific indicators",
      "format": "guidance",
      "notes": "Monitor for the techniques described in AgentCorruption (AWS Bedrock AgentCore IMDS). Refer to sources for specific IOCs and TTPs."
    }
  ]
}