Grounding with Bing Search account: intermittent 404 "ResourcePostActionFailed: Not Found" on listKeys and 400 "ResourceCreationValidateFailed" on an unchanged PUT

Since 2026-09-29 around 21:20 UTC, ARM control-plane calls on two existing Grounding with Bing Search accounts fail intermittently. The accounts are kind Bing.Grounding, SKU G1, location global, created with Bicep (Microsoft.Bing/accounts@2020-06-10) on a pay-as-you-go subscription in East US 2, one per environment. Both read back as provisioningState: Succeeded under 2020-06-10 and 2025-05-01-preview, and grounding keeps working at runtime through the existing Foundry project connections.

What fails, all from the Activity Log:

• POST .../listKeys?api-version=2020-06-10 returns 404 {"code":"ResourcePostActionFailed","message":"ResourcePostActionFailed: Not Found"}. Same identity, same request, 200 two seconds later.
• An idempotent PUT of the unchanged account returned 404 ResourceCreationFailed: Not Found once and 400 ResourceCreationValidateFailed: The resource validation failed once, with 200 for the identical body before and after.

Out of 35 calls between 2026-09-29 12:00 and 2026-09-30 12:45 UTC, 5 failed: across both accounts, two callers (a deployment service principal with Contributor on the resource group, and a user with Owner), and both operation types. Nothing changed on our side: the template is identical across successes and failures, no role assignments changed, and Microsoft.Bing is registered. Permission problems in our tenant return 403 AuthorizationFailed, which we have seen separately; these are 404 and 400 from the provider. No Service Health event was published.

Failing correlation ids (UTC):

• 2026-09-29 21:22:47, listKeys, 404: 59e7956a-1853-4743-9487-0982b54d577f
• 2026-09-29 23:05:51, PUT, 404: 98da8d79-1ea7-4cbb-b49b-6d88d11a68f5
• 2026-09-30 12:35:10, listKeys, 404: a6e09faf-bc0d-4021-8798-a1afa2fb00f7
• 2026-09-30 12:37:57, PUT, 400: 27c0781d-864d-4ba5-a475-6881f6454258
• 2026-09-30 12:42:08, listKeys, 404: 43bce5fb-0d40-4260-91a5-b737244be528

Succeeding ids for comparison: 45bf12d4-3b7b-46bb-bdb6-b84cf59cbc1b (listKeys 200 at 12:35:12, two seconds after the 12:35:10 failure, same caller) and 35fdf44d-20f5-4c0b-a13b-4e36258d183e (PUT 200 at 23:23:16 after the 23:05:51 failure, same caller, same body).

Impact: our deployment builds the Foundry GroundingWithBingSearch connection with listKeys(bing.id, '2020-06-10').key1, as in the Microsoft sample (foundry-samples, 45-basic-agent-bing), so each intermittent failure fails the whole environment deployment. We have stopped touching the account and connection during deployments as a workaround, which means we cannot rotate the key until this is resolved.

Questions:

1. Is there a known issue with the Microsoft.Bing resource provider since 2026-09-29, or a backend migration affecting Bing.Grounding accounts?
2. Has the recommended pattern of reading the key with listKeys at deployment time changed?
3. Is there anything about an account created this way that would make the provider treat it differently?