# AgentCorruption (AWS Bedrock AgentCore IMDS)

- ID: agentcorruption-bedrock-2026-10
- Reported: 2026-10-08
- Researcher: Tamir Ishay Sharbat, Lana Salameh, Dmitry Lozovoy
- Organization: Zenity Labs
- AI role: ai_agent_or_mcp_as_target (Agent / MCP as target)
- Confidence in AI involvement: not assessed
- Research / borderline: false
- Last reviewed: 2026-10-20
- Entry page: https://openattacks.ai/attacks/agentcorruption-bedrock-2026-10/
- Entry JSON: https://openattacks.ai/api/attacks/agentcorruption-bedrock-2026-10.json
- Browse title: AgentCorruption: AWS Bedrock AgentCore IMDS credential theft
- Editorial summary: Research disclosure: chat prompts to exposed AgentCore agents could reach IMDS and steal region-wide role credentials, exposing other agents, memories and secrets. AWS has tightened defaults.
- Evidence class (catalog annotation): Research demonstration
- Source-stated target environments: Cloud & SaaS; Agents & MCP
- Stage hunts and conditional defenses: https://openattacks.ai/api/defenses/agentcorruption-bedrock-2026-10.json

## Technique
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.

## Stack relevance (editorial)
Assessment: mapped. Editorial applicability reviewed against the catalog source account; not a scan or independent incident verification.
First action: Review AgentCore metadata access and least-privilege execution roles; check the current vendor guidance.
- Relationship: target; tools: bedrock; OS scope: not OS-specific / not stated; requires existing MCP connections: false.
  Relevance: The research concerns AWS Bedrock AgentCore agents with tools that can reach instance metadata.
  Condition to check: Confirm use of AgentCore, exposure to untrusted prompts and HTTP / shell tool access to metadata.
  Condition to check: Review the execution-role permissions and the defaults / mitigations described in the source.
  Evidence: https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model
A product mention alone is not a victim-side applicability mapping. Selection does not verify affected versions, permissions or compromise.

## Initial access
Prompt injection / SSRF via agent tools against exposed AgentCore agent

## AI tooling and models
- [AWS Bedrock AgentCore](https://aws.amazon.com/bedrock/agentcore/)

## Other tooling
- IMDS / IMDSv2
- [AWS Secrets Manager](https://aws.amazon.com/secrets-manager/)
- [ECR](https://aws.amazon.com/ecr/)

## Victims and sectors
Organizations deploying AgentCore agents (research disclosure; patched)

## Target geography
- unknown

## Attribution
Vulnerability research (Zenity at SecTor)

## Scale and metrics
Single prompt to one agent → account/region AgentCore blast radius (pre-fix)

## Attack flow
Catalog reconstruction of the supplied source account. Unknowns and research caveats are preserved.
### step-1: Public agent receives a prompt [access]
An exposed AWS AgentCore agent accepts attacker input.
Evidence: https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model

### step-2: Agent tools reach metadata [ai]
Prompt injection abuses HTTP or shell tools to query the instance metadata service.
Evidence: https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model

### step-3: Execution credentials are exposed [execution]
The response discloses the agent execution-role credentials.
Evidence: https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model

### step-4: Broader agent access [impact]
The role can reach other agents, memories and connected services; persistent memory hijacks are described.
Evidence: https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model

Connections:
- step-1 -> step-2
- step-2 -> step-3
- step-3 -> step-4

## Stage defense and hunting recommendations
Editorial recommendations, not validated coverage. Source-controlled query starters are drafts. Custom-schema examples require telemetry collection and field mapping; absence of a mapping means not assessed. MagicSword policy opportunities require managed Windows scope and enforcement.
### step-1: Untrusted content enters an agent session
- Visibility: Application / agent audit records
- Query origin: catalog_authored; status: draft; log scope: custom_schema
- SIGMA query download: /api/detections/agent-input.sigma.yml
- KQL query download: /api/detections/agent-input.kql
- SPL query download: /api/detections/agent-input.spl
- Hunt hypothesis: An agent ingests external content or an untrusted session identifier before a tool action or configuration change outside the approved task.
- Required telemetry: Input provenance, source type and request/session identifier; Captured tool responses or retrieved content where collection is approved; Subsequent tool calls, configuration changes and approval decisions
- Investigate: Join the ingested input to later actions by session or request ID; preserve the original content as evidence.
- Investigate: Compare the resulting action with the approved task and baseline. Receiving suspicious text alone is not successful injection.
- Limitation: Requires application instrumentation and a custom field mapping. Input exposure does not prove that the model followed instructions, executed a tool or bypassed approval. Avoid collecting sensitive prompt bodies unless authorized.
- Qualification: Collect the exposed agent input and request/session context. A received prompt alone does not establish exploitation.
- No direct MagicSword prevention claim is recorded for this stage.

### step-2: Agent tool access to cloud metadata
- Visibility: Defender-side logs where collected
- Query origin: catalog_authored; status: draft; log scope: custom_schema
- SIGMA query download: /api/detections/metadata.sigma.yml
- KQL query download: /api/detections/metadata.kql
- SPL query download: /api/detections/metadata.spl
- Hunt hypothesis: An agent tool reaches a metadata endpoint and the resulting role identity is used outside its intended task or resource scope.
- Required telemetry: Agent HTTP / shell tool audit; Workload egress / metadata access instrumentation; Role-session and downstream cloud audit
- Investigate: Look for metadata destination access from the agent's execution context and determine which credentials were available.
- Investigate: Trace that role session into agent memory, secrets and other resources; review the permissions boundary.
- Limitation: Metadata traffic may stay on the host and be absent from ordinary proxy logs. IMDS access can be legitimate; correlate the role and task.
- No direct MagicSword prevention claim is recorded for this stage.

### step-3: Unexpected cloud / inference identity use
- Visibility: Defender-side logs where collected
- Query origin: catalog_authored; status: draft; log scope: custom_schema
- SIGMA query download: /api/detections/cloud.sigma.yml
- KQL query download: /api/detections/cloud.kql
- SPL query download: /api/detections/cloud.spl
- Hunt hypothesis: A service key or role is used from an unfamiliar origin, gains broader access or causes inference usage outside its established baseline.
- Required telemetry: Cloud control-plane and identity audit; AI-service invocation / diagnostic logs when enabled; Usage, quotas, cost and key inventory
- Investigate: Pivot on key ID / principal, source network and user agent, then inspect role changes, token minting and service pivots.
- Investigate: Join inference usage and billing anomalies to the same principal; verify whether the behavior was authorized.
- Limitation: Cloud control-plane logs alone may not include model prompts or inference calls. No endpoint policy is claimed to stop cloud-only key misuse.
- No direct MagicSword prevention claim is recorded for this stage.

### step-4: Unexpected cloud / inference identity use
- Visibility: Defender-side logs where collected
- Query origin: catalog_authored; status: draft; log scope: custom_schema
- SIGMA query download: /api/detections/cloud.sigma.yml
- KQL query download: /api/detections/cloud.kql
- SPL query download: /api/detections/cloud.spl
- Hunt hypothesis: A service key or role is used from an unfamiliar origin, gains broader access or causes inference usage outside its established baseline.
- Required telemetry: Cloud control-plane and identity audit; AI-service invocation / diagnostic logs when enabled; Usage, quotas, cost and key inventory
- Investigate: Pivot on key ID / principal, source network and user agent, then inspect role changes, token minting and service pivots.
- Investigate: Join inference usage and billing anomalies to the same principal; verify whether the behavior was authorized.
- Limitation: Cloud control-plane logs alone may not include model prompts or inference calls. No endpoint policy is claimed to stop cloud-only key misuse.
- No direct MagicSword prevention claim is recorded for this stage.


## MITRE ATLAS
Mappings are catalog annotations based on described behavior, not assignments or endorsements by MITRE.
- AML.T0051.000: Direct — A prompt sent directly to the public agent redirects its tool use.
  Technique: https://atlas.mitre.org/techniques/AML.T0051.000
  Evidence: https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model
- AML.T0055: Unsecured Credentials — The metadata response exposes execution-role credentials.
  Technique: https://atlas.mitre.org/techniques/AML.T0055
  Evidence: https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model
- AML.T0080.000: Memory — The source describes persistent memory hijacks.
  Technique: https://atlas.mitre.org/techniques/AML.T0080.000
  Evidence: https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model

## Defensive notes
Least-privilege agent IAM; IMDSv2; agent's role is its blast radius

## Sources
- https://labs.zenity.io/post/agentcorruption-how-a-single-prompt-collapsed-the-entire-cloud-security-model
