Amazon SageMaker model registry now supports rollback of deployed models

Amazon SageMaker Pipelines, the first purpose built continuous integration and continuous delivery (CI/CD) service for machine learning (ML), now allows customers to rollback the latest customer approved version of ML models deployed to SageMaker hosted endpoints to a previous customer approved version of the model. Customers can now change the status of the deployed model package version in the model registry from "Approved" to "Rejected". This will trigger the model deployment CI/CD pipeline to roll back the current version of the model deployed to the SageMaker endpoint to the latest previous version of the model that was approved for deployment.

The roll back support for models in the model registry provides customers additional flexibility, enabling them to use the SageMaker model registry more comprehensively for managing models deployed to SageMaker inference endpoints. This feature is available in all regions where SageMaker projects is available. To learn more visit the documentation page.