Anthropic Claude API
Anthropic says inputs and outputs from commercial products, including the Anthropic API, are not used to train its models by default.
Training can occur when a customer explicitly opts in or submits feedback or bug reports; feedback may include the related conversation.
No opt-out is required for the stated baseline. Organisation administrators can disable thumbs-up and thumbs-down feedback in Workbench privacy settings.
API inputs and outputs are commercial customer data. Feedback excludes raw connector content but can include content copied into a conversation.
Anthropic says API inputs and outputs are deleted from backend systems within 30 days by default. Approved zero-data-retention arrangements can apply to eligible APIs; feedback-associated conversations may be retained much longer.
The commercial training disclosure explains feedback review and use, but does not provide a general human-review statement for ordinary API requests.
Foundational reference pages for teams adopting, deploying, and operating production AI with inspectable controls.