An AI tool in Foundry for analyzing documents and media to classify content, extract entities, and generate structured understanding
Hi ,
Thanks for reaching out to Microsoft Q&A.
This is typically a generic backend failure with poor surfacing of diagnostics, not a user-side validation error. In Content Understanding custom tasks, “Unexpected Error” during auto-labeling usually comes from one of these:
- Service side issue/transient failure - retry often works
- Dataset or file format issue - unsupported encoding, corrupt files, or size limits
- Label schema mismatch - recent changes to schema not synced with existing documents
- Permission/RBAC issue - especially if using shared workspace or different identity
- Region or capacity constraints - some features are still limited/preview-backed
What to do quickly?
- Retry with a small known good document (rules out data issues)
Check if schema was modified after uploading documents
- Recreate a new custom task and test auto labeling there
- Validate file types, size, encoding (UTF8 preferred)
- Check azure status + region availability
If consistent, raise a support ticket with operation timestamp + resource ID (needed since logs are not exposed)
Bottom line: if it fails even with a fresh task + simple document, it is almost certainly a backend issue or preview limitation, not your configuration.
Please 'Upvote'(Thumbs-up) and 'Accept' as answer if the reply was helpful. This will be benefitting other community members who face the same issue.