Replicate
Replicate's terms license customer data as needed to provide outputs and train customer derivative models when customers create training workloads. API predictions remove inputs, outputs, files, and logs after one hour by default.
Training or fine-tuning workloads use customer-provided data for customer derivative models. No reviewed source expressly says ordinary prediction inputs train Replicate foundation models.
Avoid or delete training workloads when not required. API predictions expire by default; web-interface predictions can be deleted manually. Use secret input types for supported sensitive parameters.
Replicate acts as a processor for customer personal information. Secret inputs are redacted from logs, outputs, and prediction metadata after reaching the model.
API prediction inputs, outputs, files, and logs are removed after one hour by default. Web-interface predictions are kept until manually deleted.
The reviewed terms and retention docs do not provide one general human-review statement.
Foundational reference pages for teams adopting, deploying, and operating production AI with inspectable controls.