MLSEC-12: Restrict access to intended legitimate consumers
Use least-privileged permissions to invoke the deployed model endpoint. For consumers who are external to the workload environment, provide access via a secure API.
Implementation plan
-
Use secure inference API endpoints - Host the model so that a consumer of the model can perform inference against it securely. Enable consumers using the API to define the relationship, restrict access to the base model, and provide monitoring of model interactions.
-
Secure inference endpoints - Only authorized parties should make inferences against the ML model. Treat inference endpoints as you would any other HTTPS API. Ensure that you follow guidance from the AWS Well-Architected Framework to provide network controls, such as restricting access to specific IP ranges, and bot control. The HTTPS requests for these API calls should be signed, so that the requester identity can be verified, and the requested data is protected in transit.
Documents
Blogs
Videos
Examples
