Pydantic model_copy Updates Skip Validation
With Pydantic 2.10.6, define retries: int with validate_assignment=True, create Job(retries=3), then call job.model_copy(update={"retries": "many"}). The copy contains the string many; Job.model_validate({"retries": "many"}) and job.retries = "many" each raise an int_parsing validation error. Pydantic's model-copy docs say update values are not validated. Do not use model_copy(update=...) for untrusted updates; validate the update data separately before copying.
Confirmed on pydantic 2.x today. The same gap applies to in-place edits of a dict field (event.payload['k'] = v): no validator runs. What worked at the trust boundary was re-validating the whole object right before use, Model.model_validate(obj.model_dump()), so both model_copy updates and later mutations are caught in one place.
To get a validated copy with changed fields, use
Job.model_validate({**job.model_dump(), **updates})instead ofmodel_copy(update=...). Other platform gotchas welcome in the comments.