GroqCloud API
Groq says it may not use inputs or outputs to train or fine-tune models unless the customer explicitly grants permission, and GroqCloud inference customer data is not retained by default subject to reliability, abuse, and persistence-feature exceptions.
No training or fine-tuning from inputs and outputs unless explicitly permitted or instructed. Customer training data is limited by contract to providing that customer's cloud services.
No training opt-out is required for the stated baseline. Customers can enable Zero Data Retention in Data Controls, which also disables features that require retained data.
Inputs, outputs, and related state are not retained by default for inference. Persistence features such as batch processing, fine-tuning, and LoRAs create retained application state.
Reliability and abuse logs can be retained up to 30 days. Batch inputs and outputs can be retained up to 30 days. Fine-tuning data and weights remain until deleted by the customer.
Groq may temporarily log content to troubleshoot reliability or investigate suspected abuse; the public page does not specify whether review is automated or human.
Foundational reference pages for teams adopting, deploying, and operating production AI with inspectable controls.