A firm's risk officer asks whether work done on a commercial Claude plan is used to train Anthropic's models. What is the documented default?

Why this is the answer

Anthropic states that by default, customer content from Claude for Work, the Claude Platform and the API is not used to train its models — which is the baseline a commercial buyer is relying on. There is no annual opt-out ritual, de-identification is not the mechanism that makes commercial content trainable, and classifier flagging concerns safety review rather than training.