Data usage and retention

“Is our data used for training?” has a different answer depending on plan, product and platform, and that is the single most common reason a Claude rollout stalls at security review.

What is documented

• Team includes no default model training on your data. That is listed as a plan inclusion, not something you have to negotiate.
• Zero data retention is separately documented for Claude Code, with its own configuration page.
• Enterprise adds custom data retention alongside audit logs, the Compliance API and a HIPAA option.
• Data residency is a separate axis: inference_geo: "us" on the Claude API pins inference to the United States at a 1.1x price multiplier, and Foundry offers an equivalent US Data Zone deployment type.

Residency is not retention

They get conflated constantly:

Question	Concept
Where is it processed?	Data residency
Is it kept afterwards?	Retention
Is it used to improve models?	Training

A US-only inference guarantee says nothing about retention. Zero retention says nothing about where processing happened. Ask all three separately.

Partner clouds have their own terms

Data handling on Bedrock and Google Cloud runs under those providers’ agreements as well as Anthropic’s. If your requirement is contractual, the contract you need may be with AWS or Google.

What this page does not claim

Per-plan training defaults outside Team, retention periods, and the exact scope of zero-retention eligibility are not reproduced here. Those are the details most likely to change and most consequential to get wrong — read them at the source and, for anything contractual, get it in writing.