Azure Custom Vision fails to load project with different error messages

Question

Azure Custom Vision fails to load project with different error messages

Iwema, Sibren 10

Hello,

I have started to use Azure Custom Vision, and it worked fine for a while until 05.03.26, when I was able to use it last without issue.

Since then, most of the times, my project fails to load with different error messages:

User's image

There are one or two more errors which I was unable to screenshot so far. One error message just states something like "Oops, something went wrong!"

However, sometimes it does load successfully, or it loads hours later.

SRILAKSHMI C 19,735 Reputation points Microsoft External Staff Moderator

2026-03-16T13:23:05.1466667+00:00
Hello Iwema, Sibren,

Welcome to Microsoft Q&A,

Thank you for confirming that you have already verified your permissions and that your account has the Cognitive Services Contributor role. Since the required access is already in place, the issue is likely related to other factors such as service limits, tenant/session validation, or portal/network behavior, rather than a missing role.

Based on the errors shown in your screenshots (for example “OperationId … not supported for CheckAccess” and “One of your Azure accounts seems to have issues”) and the fact that the project sometimes loads successfully, here are a few additional areas to check.

Verify Free-tier usage limits

If your project is using the Free (F0) tier of Azure Custom Vision, certain usage limits apply:

Maximum 5,000 training images per project

Maximum 2 projects

1 hour of training per month

10,000 predictions per month

If one of these limits is exceeded, operations in the Training API or portal may fail and sometimes appear as generic errors.

Suggested step

Open your Custom Vision Training resource in the Azure portal → Pricing tier, and consider switching to S0 (Standard) to rule out quota restrictions.

Confirm the correct Azure directory and subscription

Even when permissions are correct, the portal may show access-related errors if the session is connected to the wrong Azure AD tenant or subscription.

Please check:

The Directory (tenant) selected in the Azure portal header

The Subscription associated with the Custom Vision resource

Make sure they match the environment where the project was created.

Refresh authentication session

Intermittent errors like the ones shown can sometimes be caused by expired tokens or cached authentication data.

You may want to try Signing out of the Azure portal, Closing the browser completely, Signing in again, Opening the project in an Incognito/Private browsing session

Check browser or network-related issues

Portal requests may intermittently fail if a browser cache issue or network filtering interferes with service calls.

You can try Clearing browser cache and cookies, Testing with a different browser, Trying from a different network if possible, Ensuring your network does not block endpoints such as:

*.customvision.ai

*.cognitiveservices.azure.com

Why the project sometimes loads later

Because the project occasionally loads successfully after some time, it suggests that the project itself is intact and accessible, and the issue is more likely related to intermittent access validation or portal session behavior rather than a permanent configuration problem.

Since your Cognitive Services Contributor role is already confirmed, the issue may be caused by:

Free-tier quota limits

Tenant or subscription mismatch

Authentication/session cache issues

Browser or network filtering affecting portal requests.

I Hope this helps. Do let me know if you have any further queries.

Thank you!
SRILAKSHMI C 19,735 Reputation points Microsoft External Staff Moderator

2026-03-17T06:04:35.83+00:00

Hi Iwema, Sibren,

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!
Iwema, Sibren 10 Reputation points

2026-03-17T06:24:47.1033333+00:00

Hello @SRILAKSHMI C

I tried with Private Browser yesterday shortly after your comment, and it worked at the office via office network. I tried again just now at home, via regular browser (Firefox), private session, and Edge, and all of them failed.

I am using the Free Tier with one project, about 3000 images, unsure how much time was used for training, and no predictions yet. Tenant and subscription match.

This is expected behavior in case I go over the 1 hour training time a month? Based on my experience so far, I do not want to switch to the paying S0 tier to pay for an alpha software, but rather switch to a competitor product.
SRILAKSHMI C 19,735 Reputation points Microsoft External Staff Moderator

2026-03-17T07:34:18.9833333+00:00
Hi Iwema, Sibren,

Thanks for the detailed update.

Based on what you’ve shared, this does not look like expected behavior due to Free tier limits, and I completely understand your concern here.

About the Free (F0) tier limits

Going over the 1 hour/month training limit typically results in:

Training jobs being blocked or rejected

Clear quota-related or rate-limit messages

However, it does not normally cause:

Project load failures

Errors like “OperationId … not supported for CheckAccess”

Intermittent portal access issues

Also, your current usage:

~3000 images (within 5000 limit)

1 project (within limit)

No predictions

So based on this, quota exhaustion is unlikely to be the root cause of the project failing to load.

What your latest observation indicates

The key signal here is:

It worked once in a private session on the office network, but consistently fails now across browsers and networks

This strongly suggests the issue is not tied to browser cache or a specific network, but rather:

Intermittent backend/service issue, or

Access validation inconsistency in the Custom Vision service

The error OperationId … not supported for CheckAccess is not something expected from normal usage or quota limits, and points more toward a service-side validation issue.

Clarification,

This is not expected or intended behavior, even on the Free tier.

You should still be able to:

Open your project reliably

View and manage your data Even if training limits are reached.

Recommended next steps

Since you mentioned it worked intermittently before, this may still be transient.

Try opening the project through the Azure Portal instead of direct links.

If the issue persists Please let me know, and I will check with the backend team for further investigation.

Thank you!
Iwema, Sibren 10 Reputation points

2026-03-17T10:49:26.4866667+00:00

Dear SRILAKSHMI C,

Accessing the projects via direct link or via the Custom Vision portal (customvision.ai) results in the same behaviour.
If you specifically mean the Azure Portal (portal.azure.com), please let me know how to access the project via the Azure Portal.
SRILAKSHMI C 19,735 Reputation points Microsoft External Staff Moderator

2026-03-18T03:26:37.9566667+00:00
Hi Iwema, Sibren,

Custom Vision projects cannot be directly opened or managed from the Azure Portal (portal.azure.com). The Azure Portal only provides access to the Custom Vision resource (keys, endpoint, pricing tier, etc.), but the actual project UI (images, training, iterations) is only available via the Custom Vision portal (customvision.ai) or APIs.

So in this case, accessing via direct link vs. customvision.ai is effectively the same path and your observation confirms that the issue is not related to how the project is being accessed.

What this means for your issue

Given everything you’ve shared so far:

Within Free tier limits (~3000 images, 1 project, no predictions)

Works intermittently, then fails across browsers and networks

Error: “OperationId … not supported for CheckAccess”

This behavior is not expected, even on the Free (F0) tier.

Exceeding the 1-hour training limit would only block training operations, not prevent the project from loading or cause access-related errors like the one you're seeing.

cLikely cause

At this point, the issue is most likely related to:

A backend/service-side inconsistency in Custom Vision (access validation / CheckAccess flow), or

A transient issue affecting project metadata or authorization checks

Since this is not expected behavior and you've already ruled out client-side causes, I recommend proceeding with a deeper investigation.

Could you please share the following details:

Region of the Custom Vision resource

Approximate timestamp (with timezone) when the issue occurs

Confirmation if the issue is still reproducible right now

I can use this information to engage the backend team and investigate further.

Thank you!
Iwema, Sibren 10 Reputation points

2026-03-18T08:47:15.8066667+00:00

Hello @SRILAKSHMI C

Region: Germany West Central

Timestamps:

18.03, 09:45 “OperationId … not supported for CheckAccess”
18.03, 09:44 'Something went wrong'
17.03, 7:24ish,
16.03 08:30ish
11.03 09:00ish
09.03 13:00ish
SInce most of the times the error occurs, I can try accessing it multiple times to generate more timestamps if requested.

Issue is still reproducible.
Iwema, Sibren 10 Reputation points

2026-03-19T10:49:36.7266667+00:00

Hello @SRILAKSHMI C , as added info, I was able to use it without issue last on Thursday, 05.03.26 17:00. I did not have these errors showing up prior to this date. I noticed it first on 09.03 13:00 ish. I tried today again from circa 14:00 - 15:20 in various intervals, with continuous error message. I tried again now at 15:45, and now it loaded almost instantly successfully.
Manas R Mohanty 17,270 Reputation points Moderator

2026-04-08T07:28:13.2+00:00

Hi Iwema, Sibren,

Good day. I have gone through above conversations.

Observation per above scenario

Free quota being used here

It is working in office network and failing at home network which indicates that policy constraints might have increase recently from either your organization side or Service side.

Recommendation and follow up

We can isolate the issue by taking HAR trace in both network.

Migration note:

Please test AutoML model as alternate solution . Reference on Migration

Please help with details requested in private message to create support ticket and escalate to product group if needed

Looking forward to hearing from you.

Thank you

2 answers

Your answer

SRILAKSHMI C 19,735 Reputation points Microsoft External Staff Moderator

2026-03-17T06:04:35.83+00:00

Hi Iwema, Sibren,

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!
Iwema, Sibren 10 Reputation points

2026-03-17T06:24:47.1033333+00:00

Hello @SRILAKSHMI C

I tried with Private Browser yesterday shortly after your comment, and it worked at the office via office network. I tried again just now at home, via regular browser (Firefox), private session, and Edge, and all of them failed.

I am using the Free Tier with one project, about 3000 images, unsure how much time was used for training, and no predictions yet. Tenant and subscription match.

This is expected behavior in case I go over the 1 hour training time a month? Based on my experience so far, I do not want to switch to the paying S0 tier to pay for an alpha software, but rather switch to a competitor product.
SRILAKSHMI C 19,735 Reputation points Microsoft External Staff Moderator

2026-03-17T07:34:18.9833333+00:00

Hi Iwema, Sibren,

Thanks for the detailed update.

Based on what you’ve shared, this does not look like expected behavior due to Free tier limits, and I completely understand your concern here.

About the Free (F0) tier limits

Going over the 1 hour/month training limit typically results in:

Training jobs being blocked or rejected

Clear quota-related or rate-limit messages

However, it does not normally cause:

Project load failures

Errors like “OperationId … not supported for CheckAccess”

Intermittent portal access issues

Also, your current usage:

~3000 images (within 5000 limit)

1 project (within limit)

No predictions

So based on this, quota exhaustion is unlikely to be the root cause of the project failing to load.

What your latest observation indicates

The key signal here is:

It worked once in a private session on the office network, but consistently fails now across browsers and networks

This strongly suggests the issue is not tied to browser cache or a specific network, but rather:

Intermittent backend/service issue, or

Access validation inconsistency in the Custom Vision service

The error OperationId … not supported for CheckAccess is not something expected from normal usage or quota limits, and points more toward a service-side validation issue.

Clarification,

This is not expected or intended behavior, even on the Free tier.

You should still be able to:

Open your project reliably

View and manage your data Even if training limits are reached.

Recommended next steps

Since you mentioned it worked intermittently before, this may still be transient.

Try opening the project through the Azure Portal instead of direct links.

If the issue persists Please let me know, and I will check with the backend team for further investigation.

Thank you!
Iwema, Sibren 10 Reputation points

2026-03-17T10:49:26.4866667+00:00

Dear SRILAKSHMI C,

Accessing the projects via direct link or via the Custom Vision portal (customvision.ai) results in the same behaviour.
If you specifically mean the Azure Portal (portal.azure.com), please let me know how to access the project via the Azure Portal.
SRILAKSHMI C 19,735 Reputation points Microsoft External Staff Moderator

2026-03-18T03:26:37.9566667+00:00

Hi Iwema, Sibren,

Custom Vision projects cannot be directly opened or managed from the Azure Portal (portal.azure.com). The Azure Portal only provides access to the Custom Vision resource (keys, endpoint, pricing tier, etc.), but the actual project UI (images, training, iterations) is only available via the Custom Vision portal (customvision.ai) or APIs.

So in this case, accessing via direct link vs. customvision.ai is effectively the same path and your observation confirms that the issue is not related to how the project is being accessed.

What this means for your issue

Given everything you’ve shared so far:

Within Free tier limits (~3000 images, 1 project, no predictions)

Works intermittently, then fails across browsers and networks

Error: “OperationId … not supported for CheckAccess”

This behavior is not expected, even on the Free (F0) tier.

Exceeding the 1-hour training limit would only block training operations, not prevent the project from loading or cause access-related errors like the one you're seeing.

cLikely cause

At this point, the issue is most likely related to:

A backend/service-side inconsistency in Custom Vision (access validation / CheckAccess flow), or

A transient issue affecting project metadata or authorization checks

Since this is not expected behavior and you've already ruled out client-side causes, I recommend proceeding with a deeper investigation.

Could you please share the following details:

Region of the Custom Vision resource

Approximate timestamp (with timezone) when the issue occurs

Confirmation if the issue is still reproducible right now

I can use this information to engage the backend team and investigate further.

Thank you!
Iwema, Sibren 10 Reputation points

2026-03-18T08:47:15.8066667+00:00

Hello @SRILAKSHMI C

Region: Germany West Central

Timestamps:

18.03, 09:45 “OperationId … not supported for CheckAccess”
18.03, 09:44 'Something went wrong'
17.03, 7:24ish,
16.03 08:30ish
11.03 09:00ish
09.03 13:00ish
SInce most of the times the error occurs, I can try accessing it multiple times to generate more timestamps if requested.

Issue is still reproducible.
Iwema, Sibren 10 Reputation points

2026-03-19T10:49:36.7266667+00:00

Hello @SRILAKSHMI C , as added info, I was able to use it without issue last on Thursday, 05.03.26 17:00. I did not have these errors showing up prior to this date. I noticed it first on 09.03 13:00 ish. I tried today again from circa 14:00 - 15:20 in various intervals, with continuous error message. I tried again now at 15:45, and now it loaded almost instantly successfully.
Manas R Mohanty 17,270 Reputation points Moderator

2026-04-08T07:28:13.2+00:00

Hi Iwema, Sibren,

Good day. I have gone through above conversations.

Observation per above scenario

Free quota being used here

It is working in office network and failing at home network which indicates that policy constraints might have increase recently from either your organization side or Service side.

Recommendation and follow up

We can isolate the issue by taking HAR trace in both network.

Migration note:

Please test AutoML model as alternate solution . Reference on Migration

Please help with details requested in private message to create support ticket and escalate to product group if needed

Looking forward to hearing from you.

Thank you

Answer 1

The errors indicate an access/authorization problem when the Custom Vision portal calls the underlying training resource.

To resolve:

Verify that the Azure account being used in the Custom Vision portal still has access to the Custom Vision (Azure AI Services) resource that backs the project. If access was removed or the subscription/resource was changed, restore access or switch to an account that has access.
Ensure the account has an appropriate Azure role on that resource, such as Cognitive Services User, via Azure role-based access control (RBAC). Without this role, portal operations like loading the project and checking access can fail.
If the issue persists across browsers and machines after confirming RBAC, open a support case from the Azure portal so the service team can investigate backend issues with the specific operation IDs shown in the error.

References:

Answer 2

Hi Iwema, Sibren,

Thank you for confirming that Issue is not replicating any more at your side.

Attached case investigation summary for reference

Issue

"Contact your Azure Administrator and make sure you have access"

Validation done

On Permission

RBAC - Cognitive service custom vision contributor role exist for customer

Networking -

Temporary block on Authentication - Suggested checking firewall/proxy/HAR trace logs to find blocks on Azure URLS

Shared Service tags needed in outbound rules from NSG

Suggested Managed VNET setup

On Resource

Suggested AutoML as alternate resource.

https://learn.microsoft.com/en-us/azure/ai-services/custom-vision-service/migration-options

Customer can leverage Yolo/Tensorflow/Pytorch framework for image classification and host the model as endpoint in Azure.

We are closing temporarily and intend to go for backend investigation when issue comes back.

Appreciate your patience and inputs shared during the investigation.

Thank you