How to explain the difference of behavior of Content filters for Azure OpenAI, based on where they are deployed?

Hey,

I'm noticing an odd behavior, and would like to understand if this is a bug, or a misconfiguration.

We want to setup a content filter for our deployment of GPT4o (but same problem for GPT 3.5 Turbo).

Our content filter is pretty basic, roughly we want to setup everything at "low risk".

When we deploy our model in Sweden Central, the filter works as intended. However, if I'm deploying in US East 2, or Western Europe, the filter is not being triggered at all.

FIlter is configured exactly the same.

I reproduced this behavior on two completely azure subscription (My corporate Azure and a cutomer corporate).

One thing we noticed, when querying the model by API using the js SDK, is that when it works. in sweden, we get the results in the key "ContentFilterResult" in the payload.

But, on the "faulty" deployments, we get a new key "ContentFilterResultS", that is not a declared interface from the documentation of the SDK.

So my question is: Is this behavioral difference based on region deployment normal?