Microsoft Support Response:
We understand the impact this behavior had on your production workloads, especially given the dependency on the Anthropic Files API for your document generation processes. Based on the engineering team review, I would like to share the Root Cause Analysis (RCA) below:
What Happened?
Between August 12, 2026, and August 13, 2026, the customer experienced intermittent failures when accessing the Anthropic Files API (/anthropic/v1/files) through their Azure AI Services (AI Foundry) resource. During the impacted periods, file upload, file listing, metadata retrieval, and file download operations returned 404 Not Found responses, while model inference requests to /anthropic/v1/messages continued to operate successfully.
What Went Wrong and Why?
Based on our investigation, the behavior was associated with a broader service-side issue affecting Anthropic functionality within Azure AI Foundry. The issue occurred indirectly as part of a service outage related to a service deployment change. As a result, the Anthropic Files API capability became unavailable for affected requests during the impacted periods. The issue was not caused by customer configuration changes or deployment actions. A rollback of the affected service changes restored functionality.
How Did We Respond?
The engineering team identified the service regression and mitigated the issue by rolling back the affected deployment to a previous known good configuration. Following mitigation, Anthropic Files API functionality was restored and service behavior returned to normal operation. Internal validation and customer reports confirmed successful recovery after the rollback activities were completed.
How Are We Making Incidents Like This Less Likely or Less Impactful?
We apologize for the service interruption and understand the impact this had on your production workloads. The engineering team has addressed the underlying issue that contributed to the outage and does not expect customers to encounter the same behavior going forward. Additionally, service teams continue to improve deployment validation and service reliability processes to reduce the likelihood of similar regressions affecting Anthropic-integrated functionality in the future.