AI Foundry Connection
Does ZopNight manage AI Foundry Connection?
AI Foundry connections link projects and accounts to Azure OpenAI, AI Search, and storage; they carry no charge, but they define the AI dependency graph. ZopNight inventories every connection's category and target URL, never its credentials, because that list is the data-egress and supply-chain map of an AI estate.
Rules that fire on AI Foundry Connection
No active rule family targets AI Foundry Connection today. Rules that used to are retired, and retired rules publish no pages and fire no findings. Scheduling and permissions coverage are unaffected.
At a glance
| Field | Value |
|---|---|
| Scheduling notes | discovery and topology only. |
AI connections link Foundry projects to external resources such as Azure OpenAI accounts, AI Search services, and storage. They carry no direct cost but define the AI dependency graph.
Connections are free; the risk is what they reach
No meter runs on a connection. What a connection does is grant an AI workload a path to something else, whether a storage account full of documents, a search service, or another model provider, usually with stored credentials attached. That makes the connection list the authoritative answer to two questions security and cost teams both ask: where can this AI application send data, and which billable services can it drive load into.
What the connections enricher captures
Discovered via the AI enricher at both levels the ARM API exposes: connections registered on the Cognitive Services account itself, and connections registered on each AI Foundry project beneath it. For every connection ZopNight records the friendly name, the category (the kind of linked resource, such as AzureBlobStorage, CognitiveSearch, or AzureOpenAI), and the target endpoint URL. Credentials and secret values are never read or stored; the category-plus-target pair is what wires the AI topology between projects, model providers, and data services.
The supply-chain reading of a connection list
Three findings recur when connections get audited. Connections pointing at endpoints outside the organization’s tenant, such as an external model provider or a personal storage account, are data-egress paths nobody signed off on. Stale connections to deleted resources, which fail at runtime and mask real configuration drift. And duplicated connections to the same target scattered across projects, each with its own stored credential to rotate. None of them show up on a bill; all of them show up in an incident review.
Where connections live in Foundry
In the AI Foundry portal, the management center’s Connected resources page lists connections for the selected account or project, and is where targets and credentials are edited. The Azure portal exposes the same objects as ARM child resources under the Cognitive Services account, which is the path automated inventory takes.