azure-ml-foundry-workspace
Warn
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill provides a command example in
references/compute.mdthat uses thesetup-scripts-creation-scriptparameter to fetch and run a script from an external, untrusted source (https://raw.githubusercontent.com/myorg/scripts/setup.sh). This facilitates the execution of unverified remote code on Azure compute resources. - [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external data from Azure logs, which could contain malicious instructions designed to influence agent behavior.
- Ingestion points: The skill uses
az ml online-deployment get-logsand Log Analytics queries inreferences/troubleshooting.mdto retrieve container and system logs. - Boundary markers: The instructions lack delimiters or warnings to treat the content of logs as untrusted data, increasing the risk that the agent may follow embedded instructions.
- Capability inventory: The agent has broad access to Azure infrastructure via
azCLI, PowerShell, and Terraform, including the ability to create resources and manage identity permissions. - Sanitization: No logic is provided to filter or sanitize the log data before it is presented to the agent for analysis.
Audit Metadata