Microsoft lets AI agents change Azure resources with user rights
An agent can make each change that its user can make, and Microsoft names no approval step.
In November 2026, at Ignite, Microsoft added write tools to its Azure Resource Manager server. The server connects AI agents to Azure through the Model Context Protocol, an open standard for agent tools. Before, it only read data across more than 3,500 resource types and could start template deployments. At this time agents can make and update resources. Each tool step uses the Azure permissions of the signed-in user.
What is new
Microsoft added three tools. One makes or updates a resource. One manages resource groups. One does tasks on a resource, such as start, stop and key retrieval.
Microsoft lists no erase tool. It does not say which actions are permitted.
Proof and limits
Microsoft gave no count of the resource types that accept writes. Its documentation calls the server a preview, and says tool names can change. No Microsoft page that it published announces general availability.
Azure policy rules can stop a change if it breaks a rule, for example on region or machine type. Microsoft says these rules apply to agents and to persons.
What is not known
The only check before a change is a preview of a deployment. It does not show each change that a person makes on a resource. Microsoft describes no approval step in the server.
Insight: Cloud agents move from reading to writing
In November, Microsoft released an Azure Monitor server where 20 of 21 tools only read. Then the Azure Resource Manager server got write tools. This reads as a race to give agents write access, where the role of the user is the only limit. A sign-in as administrator gives the agent full access. A small role keeps control.
Sources
Posted